Frequently asked questions (FAQs)

Ethical and Social Implications

Do you think robots should be used as caregivers in the home?

Response by Bruce Buchanan:

Robots in the household seem like a very good thing to me, even including some of the duties of home nursing and child care. They need to be smarter and more autonomous than the Rhoomba vacuum cleaner, of course. But there are many jobs that involve monitoring, reminding, and fetching that we now hire low-paid care givers for, which could be performed by robots 24x7, with fewer problems of calling in sick, stealing valuables, and abusing the elderly.

Human caregivers are certainly capable of human empathy in ways that robots are not. But finding enough people of quality for these jobs is a problem. Why not give each of them a team of pretty-capable robots to work one-on-one with nursing home patients or kids for continuous watching & interaction, with the fully capable person checking up on the quality of care? Will robots make mistakes (e.g., allow taking, or even actively giving, the wrong medication)? Sure, but so do human caregivers. Intelligent machines can be programmed to make fewer mistakes than the worst care givers, and maybe no more than the best.

We would all like to be cared for by someone as caring as our own grandmothers, as attentive as our spouses (well, maybe more so), and as knowledgeable as our personal physicians. This just won't happen. So how can we couple the empathy of a good person with the attention and knowledge-processing of a machine? We won't really know until we try various ways.

Is AI is a threat, or could it become a threat to mankind and why?

Response by Patrick J. Hayes:

Well, I'm tempted to ask what KIND of a threat is being considered here, and a threat to WHAT exactly? Because the answer must depend on having a little more detail.

To attempt to answer your question at face value, I'd say that AI is certainly not a threat in the usual science-fiction sense. This is the imaginary scenario in which machines become (or are made) intelligent, and then as a result decide to take over, and (since once one is as smart as a human, to get much smarter is only a matter of buying some extra RAM) the machines will have the upper hand and we humans will all become slaves before we know what has hit us. Some variation of this is a widespread idea in current fiction and even among AI folk, but it is utter nonsense. It rests on a simple mistake: the identification of intelligence with ambition. Human beings evolved, and so, like all animals, we are imbued with a passionate will to dominate or succeed in competition with other living things, a will which might be paraphrased as the will to survive or the will to live. (I don't mean to imply that mankind is red in tooth and claw, only that it is unusual to find a human being who is totally disinterested in their own survival.) However, this will to live isn't identical with intelligence, and one can imagine a very intelligent artifact with a deep intellectual capacity which is totally uninterested in surviving or taking over or controlling human life or destiny. In fact all the AI systems that have ever been built have no particular interest in social or biological domination, yet some of them are undeniably very smart at what they do. Chess-players play intelligent chess, AI diagnostic systems do intelligent diagnosis, etc. etc... Just making them smarter isnt going to make them more ambitious. Intelligence, itself, is not synonymous with fitness for survival and still less is it anything particular to do with having a will to live, let along a will to dominate or control human society. So unless people deliberately set out to make something like an artificial Ghengis Kahn, I see no rational reason to suppose that our artifacts will have any particular desire to take us over, any more than the automobiles that we make are likely to start running around by themselves.

Now, this answer might seem naive, if what you mean by "AI" is the creation of something like an artifical superman, since of course human beings do have these ambitions. But thats a basic mistake about AI. The field isnt concerned with making Golems, but creating intelligent artifacts. Some AI people talk about the field as having an ultimate ambition of creating "human-level intelligence" (John McCarthy's phrase). I disagree, myself, but in any case, notice that he says human-level *intelligence*, not human-level ambition or greed or will to conquer or craftiness.

Just in terms of the likely effects of technology, I think that AI is much more likely to be a boon than a threat to humans. In many ways one can best describe AI technology as the development of what my colleague Ken Ford calls 'cognitive prostheses': systems that people can use to amplify their own intellectual capacities; that will do some of their thinking for them, just as our eyeglasses help us see things and our clothes keep us warm. We have developed this theme in some publications in more detail, but the essential point is that such tools actually *empower* people and aid in removing social barriers. To dramatize the point, its worth noting that about a hundred years ago, the ability to do rapid mental arithmetic was considered an impressive intellectual talent, and people who could do it received academic honors. Nowadays a high-school dropout at a supermarket checkout can tell the customer the total charge in a fraction of a second. A barcode scanner and a computer read-out act as a kind of mental amplifier enabling someone to perform a task that, without it, would require greater mental capacity than he could deploy unaided. True, we don't usually say that the supermarket checkout clerk is using this machinery to think with; but ask yourself: who is earning the wages, the human or the computer?

Now, I think that one could say that AI has some threatening aspects, but of a much more subtle and intellectual sort. In some ways, the scientific goal of AI might be described as understanding intelligence (including human intelligence, ultimately) in computational terms. If we do ever achieve such an understanding (we are only at the beginning of this task right now) then we will, I think, as a society, be forced to re-think our view of ourselves and our place in nature. There are some who might find this prospect threatening, as I suspect it will call into question some of our long-held views of ourselves. But this is a very tenuous kind of 'threat'.

Is the Artificial Intelligence a menace to the Human Brain in the near future?

Response by Bruce Buchanan:

I believe AI is not a menace to the human brain nor to human society. In order to survive, and to ensure survival of the planet, the human race needs more intelligence. Witness the mess we've made of controlling war, famine, greed, pollution, population growth, and myriad other global problems. Intelligent computers will be capable of thinking through the consequences of our actions and suggesting creative solutions to problems that are too complex for our politicians to solve.

Is there an AI police, a singular "governing" body for AI research, or is all work self-policed independently?

Q: Is there an AI police, a singular "governing" body for AI research, or is all work self-policed independently? Suppose I was aware of someone creating an AI without putting any forethought into ethical safeguards. Is there anyone I could report them to? Thanks in advance,

A: Response by Bruce Buchanan:

No science that I know of has a governing body to police its activities. Nuclear physics, molecular genetics, and AI are among the sciences with the most profound ethical implications, but any part of science that has the potential for benefitting humankind also has the power to do harm.

Science is based on openness. Open, honest publication of results -- and the methods by which the results were obtained -- helps assure everyone that experiments can be replicated and the results are not fabrications. Open discussion of the implications of new research also helps assure everyone that the downside risks of a line of research are tolerable in light of the potential benefits. Every scientist has an obligation to publish for these reasons.

Scientists also have an obligation to publish explanations of their work that is understandable to an educated lay public. By this means, the public and the legislators who represent them, are in a position to make informed decisions about legislation that governs some parts of science. There was recent legislation, for example, prohibiting research on stem cells that used cells not already available. The United Nations tries to limit nuclear proliferation and prohibit develop of enrichment facilities that would allow more countries to manufacture nuclear weapons. The European Union has passed strong laws limiting the use of genetically-modified foods. The current world-wide discussion of global climate change is leading to new laws restricting energy production of various kinds and promoting research on alternative energy sources.

AI is not as mature or well developed as physics and genetics so it is probably premature to ask the US Congress to pass any legislation limiting the uses of intelligent machines. But it is certainly not premature to discuss the implications of having intelligent machines. Science fiction writers have brought those implications to our attention; AI scientists continue to write both technical and non-technical articles for the lay public that provide honest assessments of the strengths and limits of AI, and of future risks and benefits. Many of these articles are collected in the Ethical & Social Issues pages. I hope you find them useful.

Games

Can a computer really bluff like a human?

Computers can definitely bluff. Although they may not bluff like humans do. Bluffing falls entirely out of the mathematics of the imperfect information game, of which poker is the quintessential example. The mathematics of game theory requires that in order to play in a fashion that cannot be exploited by other players your play should not give away what cards you hold. Bluffing (and slow-playing) are key parts of a game theoretic solution since they prevent your opponent from knowing you have a strong hand when betting nor knowing you have a weak hand because you didn't bet. Since this falls out of the mathematics, we don't even have to explicitly program bluffing into our program. The program figures out how, when, and with what frequency to bluff all on its own.

-Michael Bowling

Do you think poker computers will eventually beat human players?

Yes. For heads-up poker this is a certainty. It's possible, given enough computing power, for computers to play "perfectly", where over a long enough match (so that luck evens out) the program provably cannot lose money. However, humans will always make some mistakes, meaning the program will have an advantage. We're still quite far from the necessary computing power for perfect play, though.

The 2008 Man-Machine competition in Vancouver, though, demonstrated that computers can already play competitively with humans, even though its still far from perfect play. And the technology is evolving rapidly. In less than one year we have made significant improvements to Polaris, and I would expect that computers would be able to beat the best poker professionals in Heads-Up Limit Hold'Em within a couple of years, if not sooner.

Games of poker involving more than two players or more betting options, such as no-limit, are still a bit further out of reach.

-Michael Bowling

Does a poker robot's lack of emotion give it an advantage?

Naturally. Computers don't go on tilt. They don't get frustrated when getting cold cards or start to feel bullet-proof after a string of hot cards. One of the Polaris bots that played in the Man-Machine match in Vancouver exploited this fact with very aggressive play, although still game-theoretically sound. The humans struggled to respond to its extreme aggression. Whether the aggression spooked the players, pushed them into decisions they weren't used to making, or even simply was a style they hadn't seen before, we don't know. But certainly the players walked away from that match looking emotionally beat-up, while the program was oblivious to what just occurred.

-Michael Bowling

Doesn't creating a super-player ruin the game for humans?

No more than the existence of experts hurts regular players! People play games for the personal challenge, enjoyment, and their interest is not harmed by the fact that somewhere out there a bot could beat them. Even finding the perfect solution to a game like Checkers doesn't stop people from having fun.

Given all the important and practical applications of artificial intelligence, why do researchers spend time on games?

Games provide focused challenges for artificial intelligence, just as they do for human players. Playing games lets us tune particular skills and test different kinds of intelligence while having fun. These benefits carry over to AI research: Games give AI researchers a way of exploring different aspects of intelligence in-depth. They have a well-defined structure that can allow for clear measurement of progress. Pitting computers against humans lets us push AI and can show us surprising things about human intelligence. International competitions create a testing ground with variety and scope far greater than isolated testing.

Is poker-playing software harder to write than, say, chess-playing software? Why?

In games such as chess and chequers, players have access to all the information they need. But poker contains hidden information. How does that complicate the situation?

Poker is a very different game than Chess. Poker is a game of imperfect information, where the information a player would like to know in making their decisions (the other players' cards) is hidden. This makes poker a more challenging problem for artificial intelligence. And traditional techniques, namely search which was the centerpiece of AI in chess, cannot be readily applied in poker. As such, new techniques have been developed to handle these more challenging classes of games.

One interesting difference between high-end chess programs and high-end poker programs is when the computation happens. Polaris does the vast majority of its computation before the match starts. It uses months of computation, playing billions of hands of poker against itself, to compute a strategy for playing the game. Once in the middle of a match, it simply consults its strategy to determine how to play any given situation. Therefore its play during a match is nearly instantaneous. Chess programs do a great deal of computation during play as they search through possible moves and counter-moves. The best chess programs have to be heavily optimized for fast computation during a match. The best poker programs are heavily optimized to maximize the computing resources used before the competition. The challenges of writing software for chess and poker is quite different, rather than one being harder or easier.

-Michael Bowling

Should users of online poker sites be wary that some players could be robots?

Users of poker sites should definitely not be wary on account of our success. Our program does not play online for money and the program itself is carefully guarded for this reason. I am not an authority on the practice of programs playing on poker sites, and so can't comment on how many or the skill level of any such programs. Such practice, though, is against the terms of service of most sites and I know that sites are interested in aggressively pursuing those who violate the terms.

-Michael Bowling

The best human players try and think inside the minds of their opponents. Is this a better approach than game theory, and could computers ever do this?

Opponent modelling hasn't played a very significant role in the recent successes of poker playing programs, but it's maybe the most interesting challenge in poker. The notion of "perfect" play in game theory aims at a strategy that is guaranteed to not lose money in long matches. Opponent modelling or exploitive play, on the other hand, aims at something quite different: maximizing winnings against the current opponent. Top human players will draw conclusions about an opponent before they even sit down at the table by the way they dress, carry their chips, or their online nickname. These models are being constantly adapted after each observed decision of their opponent and inform a strong player's decisions. Perfect play may suggest that you should bluff in a certain situation 20% of the time, but if you knew your opponent was tight you could make more money by bluffing more often. Based on current state of the art, it's clear that humans do this very well and computers do not yet.

Can computers ever be able to do this? I think so. We're currently exploring some new ideas for exactly how this could be accomplished, and they look promising.

-Michael Bowling

What do you hope to accomplish through Human-Computer competitions such as the Man-Machine Poker Match or Kasparov vs Deep Blue chess tournament?

Computers are getting more powerful and their algorithms are getting more sophisticated. Games researchers have been pioneering new techniques to address many different challenges within games that are important in real-life decision making. How far have these techniques come along? Many games currently being studied for AI have a long history. The top players have devoted years of their life to playing and studying the game. Have we reached the point where computers can play these games, and deal with their particular challenges, better than the best humans which have had years of experience and benefit from decades of collective human expertise? This is what we hope to answer.

Human-vs-computer events can motivate the research. Competition is great for driving progress. Having a goal of defeating the world's best human players forces our research group to continue to innovate, pioneer new ideas, and perfect old ones.

In addition, having machines win against their human counterparts is a milestone. It demonstrates how far the technology has progressed and helps establish in what sorts of problems do computers excel, and where does more work need to be done.

-Michael Bowling

Why are events like the Man-Machine Poker Match important?

There are two reasons to hold events like the this one. One is that they motivate the research. Competition is great for driving progress. Having a goal of defeating the world's best poker players forces our research group to continue to innovate, pioneer new ideas, and perfect old ones. For example, two years ago the AAAI computer poker competition (for poker-playing programs to compete against each other) was started. In just the first two years huge strides have been made in computational techniques for solving large models of strategic interactions, and this was following ten years of no change in the state of the art.

In addition to competition pushing the research forward, this competition is an attempt to establish another milestone in the progress of artificial intelligence. Computers already play chess and backgammon better than any human, and they even play perfect checkers. But what none of these games have is the degree of uncertainty that players face in a poker game, including uncertainty in what the opponents' cards are, uncertainty in what future cards will be, and uncertainty in the opponents' playing styles. We're hoping to show that the new AI techniques that we've been developing over the last ten years can effectively handle this degree of uncertainty. This isn't just a toy problem. The techniques that we and others are developing to cope with uncertainty in poker are critical for many promising applications of artificial intelligence. Decision making problems in the real world almost always involve uncertainty. For example, when we drive a car we have to cope with uncertainty in the goals of the other drivers (where are they trying to get to?), future events (when will the light change? is that a pot hole in the road?), and others' driving styles (why did he hit his breaks? is he drunk?). Business decisions also are primarily about coping with uncertainty whether it be when participating in small scale Ebay auctions or the billion dollar high-stakes FCC spectrum auctions in the United States, the type of which is now coming to Canada. These are essentially high stakes poker matches, and the ideas that underly computer programs that can beat the best human players can also allow better decision-making in these scenarios as well.

-Michael Bowling

Why should the poker-playing public be interested in the Man-Machine Poker Tournament?

We know from [the 2007] competition that duplicate man-machine matches are both very entertaining and very educational. It is a rare opportunity to witness a world-class poker player given the opportunity to think out loud as they play. Since Polaris is not listening, watching, or even timing its opponent, the human player is free to talk about their hand, how they're playing, what they think the computer has, or trash talk all without giving away their hand. It's a bit like television poker, but the player's themselves can narrate the hand rather than announcers having to guess the player's thoughts. It is effectively a free lesson in heads-up limit hold'em from world-class players and coaches. In addition, Polaris is educational by itself. For those more mathematically minded players, Polaris takes game theoretic reasoning to its extreme. Phil Laak and Ali Eslami last year both felt they walked away having learned something from Polaris.

In addition, this is a historic opportunity to see the continuing battle of humans versus machine in games of intelligence. Computers can play a perfect game of checkers and they are superior to all humans at chess; the next round is poker. But poker is a very different game. The game is full of uncertainty that doesn't arise in chess or checkers: uncertainty in future cards, uncertainty in what your opponent holds, uncertainty in how your opponent plays. These are the reasons that computer poker programs haven't yet conquered their expert human counterparts. However, much has changed in the past two years. We believe that we are close to overtaking the best human players at heads-up limit games and this is our chance to provide. In summary, these matches are like the historic battles of Deep Blue versus Kasparov in 1996 and 1997.

-Michael Bowling

Preparing for Graduate School

What should I do to prepare for graduate school?

See what Peterson's Guide has to say about the steps you should be taking a year ahead of starting graduate school.

What undergraduate courses should I take?

Q: As far as classes go I have taken Calculus 1, Visual Basic Programming, Flash Programming, Econometrics, Statistics, and other quantitative coursework. I have never taken a course in advanced programming such as C, C++, Java and my understanding of programming right now is only at the Elementary level. I am trying to decide what would be the best use of the [remaining] time I will have....

Should I be signing up for Calculus 1 and 2 before I enter graduate school and take a course in more advanced programming such as Java or algorithms. I could take these at the Community College to refresh and expand what i have learned (And to save money).

A professor I have emailed told me that I should take English Composition or Technical writing in order to help improve my GRE score and English abilities. ... Is this a good plan of action? I feel kind of lost as to what I should be doing.

A: This is not a simple question to answer since every graduate school admissions committee has its own criteria for assessing prospective students, and each person presents a different array of skills and achievements.

It is notoriously difficult to predict success among students. Native talent and motivation are difficult to measure, but they are clearly important. You are probably a better judge of those than anyone else.

Years ago, the perceived success of students at one prestigious Computer Science graduate program was matched against the data originally collected on their applications. The single best predictor of success turned out to be the verbal GRE score -- perhaps, in part, because the variability of quantitative scores (and other mathematics-related measures) among CS students is low. Clear thinking and clear writing are probably correlated, and writing technical papers is an essential part of academic life. But there is also little doubt about the importance of mathematical concepts for the understanding of essential CS concepts and the design of sound programs.

Every AI faculty member has known excellent students who have been extremely "one-sided". Most admissions committees, however, will want to find exceptional strengths that compensate for any perceived one-sidedness.

A technical writing class sounds like it would be a good thing for you. Many graduate schools would presume you know calculus; almost all would need to know you are proficient in at least one programming language. You can look at the course requirements of most CS programs on their websites and gauge your own readiness pretty well.

Don't be discouraged by the hurdles. The satisfaction of working in AI outweighs the challenges of obtaining the skills.

Bruce Buchanan

Questions about Finding Employment

How do I write a resume?

For starters, see 10 Things to Never Put on Your Resume and readers' comments on the article as well.

Questions about school and education

Can you help me? I am doing a report about AI for school and I don't know where to begin.

We sure can! Here are some tips and suggestions.

Need a Topic? Try browsing through our collection of AI in the News articles. Other good places to look for ideas are: the AI Overview, Brief Histories & Timelines, The Future and The AI Effect.

  • How to Choose Successful Research for your PhD or Master's Degree. By Naomi Rockler-Gladen. Suite 101 (March 26, 2007). "Choosing a meaningful research topic for your dissertation or master's thesis can be a challenge. Here are some criteria to help you decide."
  • Also see: Starting the dissertation - Experts offer tips on picking a topic, conducting a lit review and narrowing your focus. By Melissa Dittmann. gradPSYCH (Volume 3, Number 1, January 2005).

Need to find something specific? Try our search engine.

Need advice about how to write a research paper?

Need advice on how to give a talk?

  • How to give a talk. By Bruce Randall Donald. "So you've been asked to give a talk in front of a seminar--or possibly in front of a much larger audience. Or maybe you've been giving lots of talks, but you wonder about how you can make your talks more effective? The purpose of this page is to present some ideas about presentation style."

Need help with the terminology? Check out our selection of Online Dictionaries, Glossaries and more.

Need to find an AI scientist to write about? You can find plenty of interesting people on these pages: Interviews, Tributes, History, Namesakes, and AI in the News.

Need help finding articles? See our links to Online Bibliographies & Digital Libraries, and various AI Journals.

Need a helpful video? Visit our Videos.

Need some software or programs ... or do you want to build a robot? Check out our Systems & Languages and Robots collections.

Responsibilities of scientists. Science is an effective method for finding the truth when scientists act responsibly. Some cardinal rules:

  • Measure and describe as accurately as possible
  • Check experimental results
  • Describe methods as fully and objectively as possible so others can replicate experiments
  • Do not fabricate or falsify evidence
  • Distinguish hypotheses that are speculation from those that are supported by evidence
  • Acknowledge the work of others, especially when using their methods or results
  • Make new findings public in timely fashion

Need an example of a student project?

  • See How to design a poster for the AAAI Student Abstract and Poster Program (with examples). From Sven Koenig, associate professor in computer science at the University of Southern California and former chair of the program. "A good poster allows someone to grasp quickly what your research is all about, and allows you to explain your ideas to them in more detail in case they are interested. It works like this: ... "
  • As part of the Intel International Science and Engineering Fair, the world’s largest high school celebration of science, the American Association for Artificial Intelligence recognizes ten high school students for their outstanding projects with an artificial intelligence component. Each winner receives a $500 cash award (joint authors shared the cash award), a one-year membership in AAAI, and a one-year subscription to AAAI’s AI Magazine for the student’s high school. " AAAI members who judged this competition were quite impressed by the caliber of work these students demonstrate," said AAAI Executive Director Carol Hamilton. "We hope this award encourages these promising young students to continue pursuing their interest in AI." See these press releases for the names of the winners and their projects: 2006, 2005, 2004, 2003.
  • See this Artificial Intelligence section, created for ThinkQuest's annual international student competition.

Useful AITopics Pages

Could you answer a few questions for a school report?

Q: What kinds of applications is artificial intelligence currently in?

A: There are literally thousands of different applications of AI in every area of industry, science, medicine, finance, defense, and government. Computers are everywhere, and no matter what they are doing it makes sense to think of software to help them work smarter, not harder. See our Applications page.

Q: What limitations are there currently on further development of artificial intelligence? (Problems that currently are trying to be solved) 

A: Two big problems, among others:

LEARNING -- Every computer program should be able to learn from its mistakes and from the preferences & behavior of its users.
REPRESENTING KNOWLEDGE – AI programs can pretty easily store and use factual information. Much of what we know, though, is information about how things work, how we can do things, what mechanisms are involved – and all this has to be integrated with specific facts. Also, the kind of common sense information that children learn in their first five years is difficult to represent and use effectively.

Q: What attempts are being made to solve some of these problems?

A: Many smart people around the world are working on these problems, with funding from government and private sources. Problems that we can identify and define precisely are much more likely to be solved than the problems we are not even aware of.

Q: Do you believe that one day robots will be able to work and live like humans?

A: They certainly will be able to work as well or better than humans. In specialized areas, computers have been shown to outperform the very best humans. Chess is the example everyone thinks of. It won’t be necessary for any single robot to be better than humans at every task, however – after all, we rarely find humans who outperform peers in many different tasks. There may be no advantage in making robots live like humans, as in the movie “AI”; it’s not even clear we will want them to.

Q: Will we one day be living among robots, if so do you think people could handle it?

A: Think about the Roomba vacuum cleaner. People already live in the same households with these specialized robots, and welcome their help. A similar commercially-available device washes floors. Put those two tasks into one robot, then begin adding other things that would be helpful – sweeping the porch, answering the phone, running to the store for eggs, whatever. Some people probably would rather do these things themselves, some may prefer paying a human. But many people would have no problem having more help with routine duties, some might even welcome having a device they could have a good conversation with.

Q: Could artificial intelligence and robots with AI be one day smarter than humans, is this possible?

A: “Smarter” means different things to different people. Computers without AI are already much faster and more accurate with arithmetic, which 300 years ago was considered to be a skill that required great intelligence. (Incidentally, people with that skill were called “computers”). Computers now can store and retrieve far more facts than humans, also a skill that people with superb memories used to sell. AI programs can now solve numerous specific problems better than people who are routinely paid to solve them. The Turbo-Tax program, for example, knows as much or more about filling out income tax forms than many accountants. Until programs can learn continuously, though, and improve their knowledge & skills by interacting with the world, most of us would say they are not as smart as three-year olds.

Q: Do you think this could happen? Do you think they could dominate humans? Should we be afraid?

A: Yes, it probably will happen. I’m not sure about the domination part, not even sure what it means. Does a cockroach feel dominated by humans? We’re probably smarter, but they seem to survive all our attempts to eradicate them. Instead of being afraid, it makes more sense to think about what are the worst sorts of things that could happen and then design safeguards so they don’t. We have much more to fear from fellow humans who practice genocide and those who knowingly sacrifice the health of our species, and the overall health of the planet, for corporate profits. We need computers with more intelligence than we have ourselves to help us think through the complex problems we humans have created.

Q: What is the most human-like case of AI have you seen or heard of?

A: The Japanese are building robots that look like people and interact with people in our world. (Look up “Aibo” on YouTube.) Their cognitive skills are not great -- that is, they’re not very smart. But then again, a lot of people are a few bricks shy of a load too and they are very human-like.

Q: What is the coolest case of AI?

A: Tough to pick just one. 

NASA’s planetary rovers are exploring unknown terrain entirely autonomously and autonomous vehicles can drive long distances on unpaved roads on their own.
The translating telephone is pretty cool, too, although not as close to daily use as other applications. With it you could carry on your half of a conversation in English with someone who knows no English and whatever they said in their own language you would hear in English. It needs to have more than ability to translate one sentence at a time from one language to another; it needs a sense of dialogue & social customs, knowledge of the world, and expectations about what people believe. And then anything else you like to share with me would be awesome. AI is contributing to our understanding of one of the big questions of all time – what is the nature of intelligence? What can be more satisfying than to work on this?

Bruce Buchanan

How do I choose an area of specialization within AI?

Q: I am a B. Tech Bioinformatics graduate. I want to pursue career in Artificial Intelligence. I even did final semester project on Neural networks in order to gain knowledge about the field. I am in processof applying to universities for Masters. But i can not decide my specialization in Artificial Intelligence. Its such a vast field. I want to master AI such that i can apply it to various fields. Kindly suggest me the specialization that is having great future scope. Please help me with this. I have been referring many papers and magazines but can not decide. Thanking you for your precious time.

Bruce Buchanan writes,

Thank you for your inquiry. As you know, there are many areas in AI, as in every other discipline. The areas you will probably make the most contributions to are very likely to be the ones that are most fun for you personally, the ones that capture your attention and match the skills and experience you have already acquired.

Ask yourself where you believe computers need to be much smarter than they are now and which of those areas seem particularly important or interesting to you. Then put as much energy as you can into making it happen.

For example, making better use of all the information and data on the web, making automobiles safer, creating substitutes for care-givers that can help older people, discovering new scientific theories or medical interventions from accumulated data. All these are areas where there is considerable activity now and in the future, but there are more.

Good luck with your studies. Learn as much as you can about the foundations of AI and Computer Science because you will be able to build on them for the next decades, while the details of hardware, languages and formats will change more rapidly.

How do I prepare for a job in AI?

There are many types of jobs and careers involving AI but two of the usual dichotomies are: academic vs industrial jobs, and research vs application jobs.

In all cases, a solid preparation in the tools of the trade is recommended. These tools include: programming languages, algorithm design, operating systems, data structures, logic & mathematics, probability theory & statistics, and the specialized topics covered in AI courses. These areas are covered in standard courses in most undergraduate and graduate Computer Science programs (see our Computer Science page - Student Resources panel), but other majors may include many of these courses as well.

Some people emphasize the cognitive science aspects of AI, for which cognitive psychology, neurobiology, and philosophy courses are also relevant.

Specialized subject areas, such as computational biology, legal reasoning, medical informatics, image understanding, mobile robots, and instructional systems, will also require specialized training in areas outside of AI. Applications-oriented work in either an academic lab or an industrial setting usually involves considerable programming. For programming to include AI, one needs a thorough knowledge of AI techniques for problem solving and knowledge representation.

Research jobs, except for implementation tasks that are well-defined, generally require advanced training in AI beyond a BS or BA degree. PhD training is recommended for anyone wanting to make AI research his or her career, and is necessary to compete for academic jobs. Understanding intelligence requires more than an ability to write programs.

How does our team get started on a Science Fair project?

Q: My name is ___ and I teach at ___ High School in ___ Scotland. I am writing to inform you that our school held the first Science Fair in Scotland last June ___ and it was extremely successful. The Science & Technology Fayre has now been held every year during June and the event has been extremely successful.

The Fair has been set up so that pupils research areas of science and technology which they find of interest to them, build a model/devise an experiment to demonstrate the science involved and finally, through multimedia applications display and present to the public.

The Fair has many aims. We plan:

  • to get pupils enthused about science and technology.
  • to allow pupils with barriers to learning, to access science.
  • to give pupils the chance to be creative in their out look to science
  • to encourage citizenship and the ability to work in groups
  • to enhance their ICT capabilities by using music, video and photograph to capture their work
  • to encourage pupils to be enterprising and make links with industry and employers.

Our intention is that every pupil that comes to the school in S1 (12 year old pupils) will take part in this event (around 180 pupils per year) the school has great ties with the local press and any sponsorship that we receive would get the relevant advertising. ... I am now working on the 2011 Fair and I have one group that wants to build a human sized animatronics robot. I think that if we work on a head this would be enough. We are really interested in trying to get this robotics head with artificial intelligience. How can we do this? We are really struggling and any help you can provide would be greatly appreciated! ... May I take the opportunity to thank you again for your time and implore you to consider our request.

-----

A: First of all, thanks to you and all the other creative teachers who devote enormous amounts of time on science fairs like yours.

Now to your specific question I have one group that wants to build a human sized animatronics robot. I think that if we work on a head this would be enough. We are really interested in trying to get this robotics head with artificial intelligience. How can we do this? We are really struggling and any help you can provide would be greatly appreciated!

As you already know, there are many aspects of intelligence and you need to define carefully which aspect(s) you want to demonstrate. You've already decided, quite appropriately, that creating a robot that moves or manipulates things in its world is more than you need to do. For example, learning is often taken to be an essential aspect of intelligent beings; so is communication in natural language. The ability to make a plan, or to revise a plan when things go wrong, is another important dimension of intelligence. Playing a game requires intelligence. So pick a dimension that seems fun and would make an engaging demonstration, but don't try to do everything all at once.

Part of picking one sort of behavior that requires intelligence is defining when you know the program has succeeded. Saying "hello world" in response to every question is not very intelligent, but it is a (very simple) question-answering program. Winning at Jeopardy against two former champions is definite, but would require more time than your students have.

You and your students will find resources on the AITopics site, www.aaai.org/aitopics. There is a lot of information here, but we hope that students can browse easily to find what they are looking for. There are also excellent AI scientists at the University of Edinburgh who may be able to help.

The mechanics of writing a program that exhibits some intelligence can be daunting. Here is where it helps a lot to start with a well-defined task, like learning to win at one specific game. It is extremely important to define the data structures that everyone on the team will use in their parts of the program. And it is important to make as much of the program as possible accessible to easy modification. For example, instead of using a number within a program to stand for a threshold for taking action, use a named variable whose value can be changed to make the program smarter.

It also helps to consider starting with a very simple version of the task and adding more and more capabilities incrementally.

I hope this helps. I'll be glad to answer additional questions, and I would be very interested in knowing how the project turns out.

best regards, Bruce Buchanan

How Important is Math for Work in AI?

Q: If you guys wouldn't mind. does math play a big role in all of your experiments?

A: According to the resources below, the answer to your question appears to be YES: 

  • Mathematics and Artificial Intelligence (BSc) The School of Computer Science at the University of Birmingham
    >> excerpt
    "Work in Artificial Intelligence has drawn heavily on that in mathematics in recent years, and a background in mathematics is increasingly important in research. Areas of AI that have particularly benefited from interaction with mathematics include machine learning, neural networks, and advanced logics. The flow has been two way. AI workers have contributed to the automated proof of mathematical theorems (some unproven by humans); to the development of new forms of logic for fuzzy and temporal reasoning; and to the development of new randomised algorithms (eg genetic algorithms) that have attracted attention from the Mathematics community.

    Specialities at Birmingham within AI include mathematical reasoning, evolutionary computing, neural networks, machine learning, and new logics. You will have the chance to study all of these, and to pursue one topic in depth for your final year project." 
  • Annals of Mathematics and Artificial Intelligence, Kluwer Academic Publishers
    Aims & Scope
    >> excerpt
    "The scope of Annals of Mathematics and Artificial Intelligence is intended to represent a wide range of topics of concern to scholars applying quantitative, combinatorial, logical, algebraic and algorithmic methods to Artificial Intelligence areas as diverse as decision support, automated deduction, reasoning, knowledge-based systems, machine learning, computer vision, robotics and planning.

    The journal is aimed at: applied logicians, algorithms and complexity researchers, Artificial Intelligence theorists and applications specialists using mathematical methods." 
  • [added 1/04] Stuart Russell on the Future of Artificial Intelligence. Ubiquity; Volume 4, Issue 43 (December 24 - January 6, 2004). "Computer scientists use a lot of mathematics, but we're interested in computation. Mechanical engineers use lots of mathematics, but they're interested in mechanisms and design. And maybe sociologists and economists will use lots of AI models, but they'll still be interested in societies and economies." 
  • [added 3/04] The Isaac Newton of logic - It was 150 years ago that George Boole published his classic The Laws of Thought, in which he outlined concepts that form the underpinnings of the modern high-speed computer. By Siobhan Roberts. The Globe and Mail (March 27, 2004; page F9). "It was 150 years ago that George Boole published his literary classic The Laws of Thought, wherein he devised a mathematical language for dealing with mental machinations of logic. It was a symbolic language of thought -- an algebra of logic (algebra is the branch of mathematics that uses letters and other general symbols to represent numbers and quantities in formulas and equations). In doing so, he provided the raw material needed for the design of the modern high-speed computer. His concepts, developed over the past century by other mathematicians but still known as 'Boolean algebra,' form the underpinnings of computer hardware, driving the circuits on computer chips. And, at a much higher level in the brain stem of computers, Boolean algebra operates the software of search engines such as Google. ... The most basic and tangible example is the machinations of Boolean searches, which operate on three logical operators: and, or, not. Algebra gets factored in to this logical equation when Boole designates a multiplication sign (x) to represent 'and,' an addition sign (+) to represent 'or,' and a subtraction sign (-) to represent 'not.'" 
  • [added 6/04] Computer Science Major. From collegeboard.com Incorporated's Career Browser. "What the Major is Like. The major in computer science begins with a liberal education and study of the necessary mathematical tools, which include calculus, discrete math, and modern algebra. Students will learn about the design, development, and analysis of hardware; they will study the organization and processing of instructions and data to perform computations by hardware devices. ... Students will study artifical intelligence in software that exhibits intelligent behavior. These programs play games, solve puzzles, recognize speech, and recognize and act on visual images. Artificial intelligence is closely connected to robotics and cognitive psychology." 
  • [added 10/04] The Age of Intelligent Machines, Chapter Three: Mathematical Roots. By Raymond Kurzweil. From Ray Kurzweil's book, The Age of Intelligent Machines, published in 1990. "The AI field was founded by mathematicians: John McCarthy, Alan Turing (1912-1954), Norbert Wiener (1894-1964), students of Alonzo Church, Claude Shannon, Marvin Minsky, and others. LISP, the primary language for academic research in artificial intelligence, was adapted from a mathematical notation designed by Stephen Kleene and Barkley Rosser, both students of Church. [fn] Mathematics has often been viewed as the ultimate formalization of our thinking process, at least of the rational side of it." 
  • [added 1/06] Math Will Rock Your World. By Stephen Baker, with Bremen Leak. Business Week Magazine & BusinessWeek Online (January 23, 2006). 
  • also see our Representation & Reasoning, Machine Learning and Vision pages
What are good schools for studying AI?

Q: What are the best undergraduate schools for me?

I have a student who loves linguistics, as well as robotics and looking into going into artificial intelligence. I wonder if anyone may suggest a list of universities which offers programs for both in the field of linguistics and A.I. Input is appreciated!

A: Thanks for asking. We have a few lists of academic programs, among other resources for students, on our Educational Resources. 
As you would guess, there is no simple answer. E.g., research-oriented universities are exhilarating for some students and overwhelming for others. Strong undergraduate programs are probably more numerous than strong graduate programs in specific areas. If your student is female, then having a female faculty mentor can make a big difference. And so on. But the material we've collected should help. Let us know if we can do more.

Q: I'm not sure this question will be interesting enough to qualify, but I've been having trouble getting this information, so here goes: I graduated from __ University last year with a degree in honors theoretical math, minors in computer science and physics. I had only a 3.3 GPA, but that comes with medical excuses. I've since taken several graduate computer science courses and done well, as well as the graduate logic sequence in our math department from __ (who is well known in his field, as I understand it), and I've managed to earn at least a couple good faculty recommendations that way. I am doing an internship, helping with research on machine learning image enhancement algorithms for the __ , at __ at the moment, and I've been part of a Machine Learning seminar at __ for the past few quarters. I'll be taking the GRE this summer (I was a national merit scholar, I *hope* I'll do fairly well on that), and the computer science gre this fall. I'll be applying this fall/winter to graduate school. I just want to find a few graduate schools with decently interesting Machine Learning/AI programs (I really like decision making, game theory... something in between being really applied and really abstract).... that I might actually be able to get into (I'm pretty sure CMU is out of the question). Do such even exist? Where should I look? I've asked around at __, but none of the faculty seem to pay much attention to other schools' graduate programs...... and I really don't know where to go next to get advice on this (most ranking systems fall short of providing this detailed information).Thanks for your time... [C: 7/29/04]

A: It sounds like you have a strong interest in machine learning and AI, and I certainly encourage you to follow this interest - it's a great area! If you're looking into graduate programs, there are many that have strong machine learning research groups. I'd suggest you look through the recent AAAI, ICML, and NIPS conference proceedings and see for yourself which universities the research papers are coming from -- that's probably the single best way to find out who's doing what. \\Hope that's helpful. By the way, graduate admissions committees often look more at your recommendation letters and GRE scores than your undergrad grades. Good luck with your applications to grad school! [T: 8/2/04] 

Also see our Educational Resources, especially graduate school resources.

What courses should I take in school?

Q. I live in Australia. I am planning on entering a career in robotics and artificial intelligence and was unsure if my course selection for university next year is suitable.

Because the fields of robotics and artificial intelligence are so diverse, I thought that you would appreciate hearing from a few experts.

The first few responses are part of the Scientific American Frontiers "Cool Careers in Science" web site.

"If I'm a student thinking about a career designing and building robots, what can I do now to prepare?"

Manuela Veloso: Do well what you are interested in. Get a solid mathematical and engineering background. Biology and cognitive science are also very relevant for building robots. Get a broad view of what you think robots can be useful for.

Manuela Veloso is a professor at Carnegie Mellon University who designs soccer-playing robots that have won international RoboCup competitions.

"What educational background do you need to design robots?

Maja Mataric: Again, we need to be clear on what you mean by "designing robots." If you mean designing robot bodies, that requires knowledge of mechanical and electrical engineering, and knowing computer science also helps, but is not necessary. If you mean designing robot minds and behaviors, then you need a background in computer science, and electrical engineering also helps. Finally, if you are interested in animal-inspired robotics, it is very helpful to study biology, ethology (the study of animal behavior in nature) and cognitive science.

"If I'm a student thinking about a career in designing and building robots, what can I do now to prepare?"

Maja Mataric: For those interested in designing and building robots, there are three career paths: academic research at a university (what I do), working for a company (these companies build robots and sell them to universities), and working for NASA (the main non-academic user of robotics today). In the future, as robotics becomes a larger part of everyday life, there will be more companies and more career possibilities. To prepare for doing robotics, see my answer above, to the question about educational backgrounds.

 Maja Mataric is working on developing the next generation of intelligent robots.

"If I'm a student thinking about a career designing and building robots, what can I do now to prepare?"

Roger D. Quinn: Read, study and enjoy science and math. Tinker with mechanical and electronic devices. Learn how they operate and why.

Roger D. Quinn  has teamed up with biologist Roy Ritzmann to design and build a robot that imitates the cockroach, an insect with superior locomotion.

The next response is from one of the many interviews with Rodney Brooks (who, by the way, grew up in Adelaide, Australia). Ask The Scientists: Almost Human. "Is it possible to create a computer that mimics a human being? That's the goal of Rodney Brooks, who hopes his robot Cog will have the intelligence of a six-month-old human baby. Following the [Scientific American] Frontiers special Robots Alive!, Rodney answered viewers' questions in an Ask the Scientists panel." Here is a sample:

"I am interested in a career like yours, designing and building robots. What courses would I have to take in college? Do you have any other helpful information to help me get started in the field of robotics?"

Rodney Brooks: In high school (and college) make sure you take plenty of math -- it is the foundation for all good engineering. In college you could major in mechanical engineering, electrical engineering, aeronautics and astronautics, or in computer science. Robotics is very interdisciplinary and so except at a very few colleges there is not a major that is exactly fitted to robotics. While an undergraduate see if there are any robotics projects on your campus and see if there is any way to become an undergraduate research assistant on the project. Hands-on experience is the best way to learn about all the interdisciplinary aspects of robotics.

Whatever major you take, try to at least get the core courses in each of mechanical and electrical engineering, and in computer science. If majoring in mechanical or electrical engineering take some control theory courses. In computer science (or engineering) take courses in microprocessor control.

From An Interview with Artificial Intelligence expert Ruth Aylett. The Science Teacher (the National Science Teachers Association's journal for high school science teachers). January 2003, page 52.

"What educational background is needed to design robots?"

Ruth Aylett: Robots are composed of several systems working together: the controller is the robot's brain, which controls its movements; the body is the robot's physical appearance related to the job it performs; mobility, or how the robot moves, depends on the job it performs (for example, a robot uses propellers and rudders in the water, and legs or wheels on land); power is used to fuel the robot (electric solar cells are one example, such as the solar-powered robots described here); sensors provide signals to give robots a perceptual understanding of their environment so they can alter their behavior based on that information; and tools are unique to the task the robot performs. Just as robots are made of several systems, the field of AI requires a collaboration of many different disciplines to be successful. Engineering is clearly useful, but I know people who have a background in biology, psychology, physics, and computer science. What's most important is a willingness to learn a lot of new things from a variety of disciplines.

Georgia Tech's Ronald Arkin. (September 12, 2005). "Technology Research News Editor Eric Smalley carried out an email conversation with Georgia Institute of Technology professor Ronald C. Arkin in August of 2005 that covered the economics of labor, making robots as reliable as cars, getting robots to trust people, biorobotics, finding the boundaries of intimate relationships with robots, how much to let robots manipulate people, giving robots a conscience, robots as humane soldiers and The Butlerian Jihad."

"What's the most important piece of advice you can give to a college student who shows interest in science and technology?"

Ronald Arkin: Pay attention to basics. Defer gratification until you have mastered the fundamentals of mathematics, physics, and the other disciplines. Also pay attention to interdisciplinary studies - there's much to be learned by being a generalist. Also watch and learn from your more senior counterparts and find good role models.

Also see:

    • Cynthia Breazeal's answer to the question: "I'm just starting my B.S. in Computer Science. What educational path should I take to get into social robotics and AI [artificial intelligence]?" From the Ask the Expert portion of NOVA scienceNOW's Profile of Cynthia Breazeal (November 2006).

Bruce Buchanan adds,

I believe the subjects that will give you the best preparation are mathematics and science. Ordinarily, you would not begin to specialize in AI until university, and only then after taking several courses in computer science, which mostly assume a good working knowledge of mathematics. You will want to be very familiar with at least one programming language and one operating system before you start specializing.

One thing to consider is why you would like to make computers smarter. Smart computers can be used, for example, to help people stay healthy, to make transportation safer, to provide more relevant information from the web, to crete household robots, to explore outer space, to assist scientists with theory formation, and to make businesses more efficient & profitable. If there is a particular kind of application that you feel interests you a lot, then you will want to be sure you take courses in that area as well as in computer science.

I hope this helps. Good luck with your studies, and be sure you enjoy them.

What GCSE options should I select?

Q: I am 14 years old and live in England. I have to choose my options soon, and am interested in a career in Artificial Intelligence. As these GCSE options will affect my future career choices, I was hoping you could provide me with the necessary information so that I can choose the right subjects. ...

A: Thanks for your interest. We are very happy to help.

You might start by perusing the AITopics information portal. The Educational Resources page contains a lot of relevant background information.

Also, some of the other FAQs are relevant.

To answer your question more directly, I believe the subjects that will give you the best preparation are mathematics and science. Ordinarily, you would not begin to specialize in AI until university, and only then after taking several courses in computer science, which mostly assume a good working knowledge of mathematics. You will want to be very familiar with at least one programming language and one operating system before you start specializing.

One thing to consider is why you would like to make computers smarter. Smart computers can be used, for example, to help people stay healthy, to make transportation safer, to provide more relevant information from the web, to crete household robots, to explore outer space, to assist scientists with theory formation, and to make businesses more efficient & profitable. If there is a particular kind of application that you feel interests you a lot, then you will want to be sure you take courses in that area as well as in computer science.

I hope this helps. Good luck with your studies, and be sure you enjoy them.

Bruce Buchanan

What is the correct way to cite AITopics in a school report as an electronic resource, Internet source, or web page?

There are many acceptable citation styles to choose from unless, of course, your teacher has already specified one. Which style is right for you may depend upon factors such as whether you are in high school or graduate school, and whether the course is one in the humanities or the sciences.

When Mike Hamilton, the webmaster of the AAAI site, was asked how to cite AAAI & AITopics pages and resources for a school report, he replied:

When you document sources from the World Wide Web (WWW), the Modern Language Association (MLA) suggests that your Works Cited entries contain as many items from the following list as are relevant and available: 

  • Name of the author, editor, or compiler, alphabetized by last name and followed by any appropriate abbreviations, such as ed.;

If you are citing pages from the AITopics area of the website, the editor's name is Bruce Buchanan. If you are citing from a "call for papers" or other document listing a program chair or cochair, the author of the document would be the chair or cochair. If you are citing a paper, the author would be the stated author of the paper. For all other areas of the web site, the author should be considered the Association for the Advancement of Artificial Intelligence.

  • Title of an article, or other short work within a scholarly project, database, or periodical, in quotation marks;
  • Title of a book, in italics or underlined;
  • Name of the editor, compiler, or translator of a book (if applicable and if not cited earlier), preceded by any appropriate abbreviation, such as ed.;
  • Publication information for any print version;
  • Title of the scholarly project, database, periodical, or professional or personal site (in italics or underlined), or, for a professional or personal site with no title, a description such as home page2;
  • Name of the editor of a scholarly project or database (if known);
  • Version number (if not part of the title) or, for a journal, the volume, issue, or other identifying number;
  • Date of electronic publication or posting or latest update, whichever is most recent (if known)

The AITopics web site is usually updated on a daily basis. You should therefore cite the date that you used the site.

  • Name of any institution or organization sponsoring or associated with the Web site;

The site is published by the 'Association for the Advancement of Artificial Intelligence', which is located in Menlo Park, California.

  • Date you accessed the source URL (in angle brackets);

Although no single entry will contain all fourteen items of information, all Works Cited entries for Web sources contain the following basic information:

Online document -

  • Author's name (last name first)
  • Document title
  • Date of Internet publication
  • Date of access
  • URL

[March 2005; updated March 2007]

What programming languages should I know?

Q. I am applying for a Masters course in AI, but in the interim I would like to get a head start in AI programming. Can you suggest which programming language(s) I should learn?

Bruce Buchanan writes,

A good understanding of computer science is important for AI, as well as facility with at least one programming language and operating system. Generally, undergraduate classes in algorithms and data structures provide a good introduction to fundamental concepts. Any programming language can be used, but an interpreted language generally makes program development easier. Programming for AI has traditionally been done in LISP or Prolog, but any language with strong symbolic-processing features can be used. An object-oriented language like C++ is a reasonable alternative to LISP or Prolog. Python and Java are also used. In developing and testing new ideas, which can take weeks, months or years, the speed of implementation is far more important than the run-time speed of the program, which is usually measured in seconds or minutes.

Joshua Eckroth adds,

You may want to look at Peter Norvig's analysis of the changing popularity of Lisp, Java, et al. in AI research. Generally speaking, Lisp is becoming less popular and Java is becoming more popular.

Where can I find images of historical figures or inventions I can use for a presentation?

The Computer History Museum has an image-filled timeline exhibit and their copyright policy is quite student friendly!

Some other timelines, as well as an assortment of historical resources, can be found on our History collection.

Because copyright is a major concern, you may want to look for images that are in the public domain. (See generally: Stanford University Libraries and Academic Information Resources's Copyright & Fair Use Center > Copyright & Fair Use Overview  > The Public Domain.) Since U.S. government works are in the public domain, you may also look for collections such as NASA's Image Exchange. As noted in that site, the images are not copyrighted.

Questions about the field of AI

Are the computers today powerful enough for Artificial Intelligence?

Q. After strolling around on your website the other day, I came across a frequently asked questions page on Machine Learning. One of the questions was, are the computers today powerful enough for Artificial Intelligence? And the answer was, I believe the computers of 30 years ago were powerful enough if only we knew how to program them. Which leads me to my question. In current research, is AI research being programmed on top of exisiting operating systems, and basically are programs running on top of other programs? I've started work on a kernel, which IS the AI program, not a program on top of a kernel. Considering I dont know anything about AI, I've come up with the theory that, in order for AI to work cleanly, it must have DIRECT access to the hardware, not have to make system calls to access hardware and memory and so forth. But I dont want to start working unless I know for sure that the research currently being conducted isnt already based on this.

Any Reply Appreciated,
J.R. 

A: As you've noted, AI research is largely based on existing hardware and operating systems. Since the mechanisms for achieving intelligent behavior are not at all well understood, researchers need experimental environments that are easy to work in. They (we) assume that once some of the mechanisms are worked out at a conceptual level it will be possible to optimize them by mapping them into hardware or systems capabilities.

An example from the early history of AI is McCarthy's mapping the powerful idea of linked lists into the Lisp language, and then people at Xerox and TI building special-purpose Lisp machines with those constructs in the hardware. It gained speed, but the conceptual advances do not seem to me to be that great. Another example is Danny Hillis' construction of the Connection Machine, with the concepts of neural networks mapped into massively parallel machines. Both machines provided nifty platforms that ran AI programs faster, but they did not seem to solve conceptual problems.

The main advantage of avoiding system calls would seem to be speed. We're more hobbled by lack of ideas than slowness of operation, I believe. There may well be AI researchers who disagree, but I don't know who they are.

Good question.

Bruce Buchanan (10/6/03)

What are the applications of AI?

Q. What kinds of applications make important use of artificial intelligence?

There are literally thousands of applications of AI in every area of industry, science, medicine, finance, defense, and government. Computers are everywhere, and no matter what they are doing it makes sense to think of software to help them work smarter, not harder. See our Applications topic overview, including its suptopics and the Innovative Applications of Artificial Intelligence (IAAI) papers for a wide range of applications. AI in the News is another great resource for recent research and product announcements that incorporate AI.

What is Artificial Intelligence?
  • What is AI? - Professor John McCarthy received many email inquiries about what artificial intelligence is all about. This was his first attempt at answering them on a layman's level or beginning student's level.
  • See Interviews - because it's not uncommon for questions such as "How do you define AI?" and "Where do you think AI will be in 10 years?" to come up in the course of an interview.
  • AI FAQ Collection from Pamela McCorduck, author of Machines Who Think: 25th anniversary edition. Natick, MA: A K Peters, Ltd., 2004.
What would be a good direction to orient my career in AI?

As you know, there are many areas in AI, as in every other discipline. The areas you will probably make the most contributions to are very likely to be the ones that are most fun for you personally, the ones that capture your attention and match the skills and experience you have already acquired.

Ask yourself where you believe computers need to be much smarter than they are now and which of those areas seem particularly important or interesting to you. Then put as much energy as you can into making it happen.

For example, making better use of all the information and data on the web, making automobiles safer, creating substitutes for care-givers that can help older people, discovering new scientific theories or medical interventions from accumulated data. All these are areas where there is considerable activity now and in the future, but there are more.

Where can I find information about "old" AI programs, systems, and projects?

Q:Do you know where I can find information about "old" AI programs, systems, and projects?

A:A good source for this type of information is IEEE's Annals of the History of Computing. And don't forget to check out oral histories, such as those in the The Babbage Institute's oral history collection, for they are an excellent source of anecdotal information.

Where else can I find other AI FAQs?

Artificial Intelligence FAQs. Easy access to the collection of FAQs that moved from CMU to UCLA. Maintained by Amit Dubey and Ric Crabbe; written by Ric Crabbe, Amit Dubey, and Mark Kantrowitz.

Individual Topics

  1. Artificial Life
  2. Expert System Shells
  3. Fuzzy Logic
  4. Genetic Algorithms (six parts)
  5. Medical Informatics FAQs. Maintained by Aamir Zakaria, M.D.
  6. Natural Language Processing FAQ. Maintained by Dragomir R. Radev. Dept. of Computer Science, Columbia University.
  7. Neural Nets (seven parts)
  8. Neural Networks FAQs. Maintained by Warren S. Sarle.
  9. Real-Time FAQs. From the comp.realtime newsgroup.
  10. Robotics miniFAQ for Beginners. By John Piccirillo. "This miniFAQ is intended as a source to find the answers to questions most often posed by beginners: Where do I start?; What do I need to know?; Where can I find information and supplies?; and, Where do I go for help?"
  11. Robotics FAQ. NASA Robotics Education Project answers questions from students and educators. Example: "I am a 9th grade science teacher. I am interested in incorporating a robotic lab into one of my lesson plans. Are there groups that come to high schools to perform demonstrations?"
  12. Speech FAQs. Topics covered include: neral information, signal processing, speech coding and compression, natural language processing, speech synthesis, and speech recognition. Older CMU site, but still relevant.

Have your own question? Ask us!