Technology can play a part in complementing human space travel by freeing up astronauts from certain tasks to allow them to focus on more important research. "[AI could be used to] automate tedious tasks," explains Dr Kiri Wagstaff, a computer and planetary scientist in the US who previously worked at Nasa's Jet Propulsion Laboratory in California. "On the surface of a planet, humans get tired and lose focus, but machines won't." The challenge is that vast amounts of power are needed to operate systems like large language models (LLM), which can understand and generate human language by processing vast amounts of text data. "We are not at the point of being able to run an LLM on a Mars rover," says Dr Wagstaff.

A team of researchers at the University of Toronto has found a way to enhance the visual perception of robotic systems by coupling two different types of neural networks. The innovation could help autonomous vehicles navigate busy streets or enable medical robots to work effectively in crowded hospital hallways. "What tends to happen in our field is that when systems don't perform as expected, the designers make the networks bigger--they add more parameters," says Jonathan Kelly, an assistant professor at the University of Toronto Institute for Aerospace Studies in the Faculty of Applied Science & Engineering. "What we've done instead is to carefully study how the pieces should fit together. Specifically, we investigated how two pieces of the motion estimation problem--accurate perception of depth and motion--can be joined together in a robust way."

Modelling functions of sets, or equivalently, permutation-invariant functions, is a long-standing challenge in machine learning. Deep Sets is a popular method which is known to be a universal approximator for continuous set functions. We provide a theoretical analysis of Deep Sets which shows that this universal approximation property is only guaranteed if the model's latent space is sufficiently high-dimensional. If the latent space is even one dimension lower than necessary, there exist piecewise-affine functions for which Deep Sets performs no better than a na\"ive constant baseline, as judged by worst-case error. Deep Sets may be viewed as the most efficient incarnation of the Janossy pooling paradigm. We identify this paradigm as encompassing most currently popular set-learning methods. Based on this connection, we discuss the implications of our results for set learning more broadly, and identify some open questions on the universality of Janossy pooling in general.

For the past 15 years, NASA's Mars Reconnaissance Orbiter has been doing laps around the Red Planet studying its climate and geology. Each day, the orbiter sends back a treasure trove of images and other sensor data that NASA scientists have used to scout for safe landing sites for rovers and to understand the distribution of water ice on the planet. Of particular interest to scientists are the orbiter's crater photos, which can provide a window into the planet's deep history. NASA engineers are still working on a mission to return samples from Mars; without the rocks that will help them calibrate remote satellite data with conditions on the surface, they must do a lot of educated guesswork when it comes to determining each crater's age and composition. For now, they need other ways to tease out that information.

Any researcher who's focused on applying machine learning to real-world problems has likely received a response like this one: "The authors present a solution for an original and highly motivating problem, but it is an application and the significance seems limited for the machine-learning community." These words are straight from a review I received for a paper I submitted to the NeurIPS (Neural Information Processing Systems) conference, a top venue for machine-learning research. I've seen the refrain time and again in reviews of papers where my coauthors and I presented a method motivated by an application, and I've heard similar stories from countless others. This makes me wonder: If the community feels that aiming to solve high-impact real-world problems with machine learning is of limited significance, then what are we trying to achieve? The goal of artificial intelligence (pdf) is to push forward the frontier of machine intelligence. In the field of machine learning, a novel development usually means a new algorithm or procedure, or--in the case of deep learning--a new network architecture.

Overall, automation may be more likely to change jobs than to destroy them. Get ready for a new career as an "empathy trainer", "explainability strategist," or perhaps as an "artificial intelligence safety engineer". As new technologies cut a swathe through traditional white collar jobs, these are some of the new roles that could rise up in their place, according to a new book by Accenture chief technology officer Paul Daugherty and fellow researcher Jim Wilson. It is a topsy-turvy world perhaps, where economic success is defined as ensuring there is enough work to keep everyone gainfully employed. But that hasn't stopped people through the ages fretting that new technology may leave us all paupers.

I report on the tenth annual AAAI/ SIGART Doctoral Consortium, colocated with the National Conference on Artificial Intelligence (AAAI-05). I discuss highlights of the consortium and include pointers to the consortium web site. The consortium is consistently praised by students as an exceptionally educational and valuable part of the conference. Sixteen students (5 women and 11 men) presented their work, selected from an extremely competitive applicant pool of 42 submissions. Their research represents a variety of subfields of AI, including planning, natural language, probabilistic reasoning, and machine learning.

Artificial intelligence is seeping into every nook and cranny of modern life. AI might tag your friends in photos on Facebook or choose what you see on Instagram, but materials scientists and NASA researchers are also beginning to use the technology for scientific discovery and space exploration. But there's a core problem with this technology, whether it's being used in social media or for the Mars rover: The programmers that built it don't know why AI makes one decision over another. Modern artificial intelligence is still new. Big tech companies have only ramped up investment and research in the last five years, after a decades-old theory was shown to finally work in 2012.

Artificial intelligence is seeping into every nook and cranny of modern life. AI might tag your friends in photos on Facebook or choose what you see on Instagram, but materials scientists and NASA researchers are also beginning to use the technology for scientific discovery and space exploration. But there's a core problem with this technology, whether it's being used in social media or for the Mars rover: The programmers that built it don't know why AI makes one decision over another. Modern artificial intelligence is still new. Big tech companies have only ramped up investment and research in the last five years, after a decades-old theory was shown to finally work in 2012.

This article originally appeared in the July 3, 2017 issue of SpaceNews magazine. Mars 2020 is an ambitious mission. NASA plans to gather 20 rock cores and soil samples within 1.25 Mars years, or about 28 Earth months -- a task that would be impossible without artificial intelligence because the rover would waste too much time waiting for instructions. It currently takes the Mars Science Laboratory team at NASA's Jet Propulsion Laboratory eight hours to plan daily activities for the Curiosity rover before sending instructions through NASA's over-subscribed Deep Space Network. Program managers tell the rover when to wake up, how long to warm up its instruments and how to steer clear of rocks that damage its already beat-up wheels.