This reminds me of the robot someone built to try to replace retiring bridge inspectors. It was programmed to do what they did in terms of tapping on a beam, etc., but was also given some senses that humans lack, for example lasers. The robot learned by “apprenticing” with one inspector, then moved on to another. At some point it started giving better results than the human inspectors because it had started using lasers etc. to examine a bridge structure while doing other tests it had learned from humans. It wasn’t told what to do with its lasers, just given a brief amount of data on what they could do, and it analyzed patterns it was observing and ‘guessed’ that applying laser capability in some situations would give more and/or better data.
Not being connected to the internet or any other source, it hasn’t asked for additional instruments it judges might be useful, but I have little doubt that if it had access to information on what instruments scientists and engineers use for various purposes that it both could and would!
I don’t think this is entirely correct. If this uses machine learning as I understand it, you would have to feed the program a ton of data from lasers and then tell it which measurements are “bad” and which are “good”. The AI has to be trained with the data and input from humans. It doesn’t just look at the data and go from there, nor does it request a certain type of data. If any other form of instrumentation is going to be used it will have to go through the same process.
In fact, those annoying Captcha things you have to click on to prove you are a human are being used to train AI’s, as another example. It is learning from humans to pick out buses in a picture.
It had built up a database of what constituted “good” via all the ‘training’ with the human bridge inspectors, so I presume it just built up its own reference and matched laser results with the other results.
The robot was actually featured on a TV show some years back; it was fascinating to see it using a hammer to copy what the human inspector was doing to ‘listen’ to the metal (hammers is the only instrument set I remember the robot having, but there were several others). The robot actually had an advantage when they equipped it with remote sensors so it could tell what a hammer stroke sounded like not just from up close but at intervals along a bridge.