Forget the more recent TV show, which ended up so frustratingly opaque as to render it pointless. The most fun version of Westworld is Michael Crichton's original movie. A robot cowboy comes to life and goes nuts in a theme park. What more could anyone need? Eleven years on, it's still hard to believe this film exists.

Edge cases are those instances of a particularly rare event in a certain environment. In the UK, the classic example is the instance of a particular type of wet leaves on the railway lines causing the trains to stop running. In the arena of driverless cars, there are many examples such as roadworks or car accidents. In language, jokes make edge cases. So much of AI these days works on the back of big data.

They include Russia, China and Iran. The signatories are obliged to bring in laws dictating how AI may and may not be used. The agreement aims to remove discrimination from AI and limit mass surveillance of people going about their daily business. But there are no sanctions for rule-breakers. One could, therefore, reasonably argue that the agreement allows countries to look as if they are signing up to a worthy cause whilst in practice they do not change anything.

Machine learning has achieved amazing things, but we all know its limitations. For example, the very latest natural language processing tool still exhibits no common sense. Well, here's one reason: because machine learning is designed to answer the question, what, but not the question, why. It thrives on correlation and ignores causation. Gary Marcus rightly observes that this gap in AI research needs filling.

Neuromorphic computing is seen as potentially fertile territory in which to mimic the workings of the brain. It has taken a significant step forwards with the development of the SpiNNaker supercomputer at the University of Manchester. There is still a way to go. SpiNNaker can simulate in the thousands of neurons. The brain runs to billions of neurons.

Whilst the premises upon which the story rests are somewhat improbable, the exploration of the tensions that might arise upon introducing artificial general intelligence to a household are gripping. AGI for these purposes takes the form of a human-level-intelligent robot called Adam. In practice, of course, such an AI will always think faster than humans and have access to vastly more data. It is not entirely clear, although Alan Turing, who makes a surprise appearance, thinks he does. The truth is, we cannot be sure that anyone other than ourselves possesses consciousness; we can only surmise.

I have written before about the Luddites and how autonomous intelligence might unleash a neo-Luddite movement. It would appear that day may have indeed arrived. There are reports from Arizona that autonomous vehicles are being vandalized. Since 2017, there have been almost two dozen acts of sabotage against driverless vehicles operated by Waymo in the city of Chandler near Phoenix. One unidentified assailant slashed a van's tires, other people have thrown rocks at passing driverless vans.

The 2017 conference featured a tight group of papers on the back of an effective review process. The conference addresses the question of how to represent data to the machine learning algorithms that process it.

Corti, an AI start-up, has developed an algorithm that can spot imminent heart failure in people calling up emergency services. At the moment, the algorithm is just working on what the callers are saying. This involves the actual words, i.e., natural language processing. It has been trained on emergency call logs going back years. But there is no reason in principle why it could not analyse intonation and breathing patterns from the calls too.

The folks at DeepMind are seeking to contribute to AI safety. They have designed a 2D testing environment for algorithms. The environment does not purport to cover all possible AI safety problems. For example, interpretability, multi-agent and formal verification safety problems are not covered. But a decent number is covered.