Rise of the machines
I enjoyed Jesse Crosse’s article about artificial intelligence in autonomous cars (Under the skin, 24 February).
Computer-controlled machines work accurately almost all of the time, but when things do go wrong, they tend to do so catastrophically. This usually can be traced back to something the coders overlooked.
As applications become more general, there’s greater scope for such oversight. This is where machine learning comes in. The ‘intelligence’ to carry out a task is formed by feeding the software a large number of examples of situations and their solutions, then putting it to work on new situations within the scope of these ‘learning sets’. The problem is that these are precisely defined areas. Roads, especially in urban environments, are nothing like that: the scope for variety is practically endless. Most of the time, the AI will perform beautifully, but when a ‘surprise’ occurs, as it most certainly will, it might not be able to handle it.
To make matters worse, current car AI has a one-second delay built in to give it time to decide if an apparent emergency is a false interpretation. In an emergency, one second is a long time, perhaps long enough to take the AI past the point where a solution is possible. An experienced human driver would react immediately, perhaps with a better outcome.
Currently, this isn’t a technology to inspire confidence, yet already there’s talk of Ai-controlled flying taxis – so all of the above with an added dimension. Are they mad? If it comes to pass soon, the best advice may be to stay out of cities when you can and never travel in one, because what goes up must indeed come down – and not necessarily under control. Jim Robertson
East Kilbride, Lanarkshire