Interesting and topical. First a counter-query to clarify things though; "making sense of the world" in what sense?

Making sense of the human world? biological world? physical world?

.

Thanks for starting to unpack the question, Anand,

I left the phrase "making sense of the world" deliberately vague, to avoid prejudicing possible interpretations, leaving the question wide open.

One answer is the sense made of all human experience by scientific approaches: physics and astronomy, biology, mathematics, psychology, economics… all of it.

If we can regard AI systems as potential “autonomous research assistants”, will they be able to discover and communicate new ways of structuring the data they acquire from their explorations, of the world that they can access, which might enhance our understanding of ways to structure our own experience?

Imagine, for example, the Internet of Things including not only access to static sensors (webcams, microphones, telescopes and ranges of scientific instruments) but also drone-powered sensors…

This could provide an exploratory AI system with the technological equivalent of a sensori-motor system – out of which could emerge whole new machine-based interpretations of their sense-data, just as valid as our own data…

Would an autonomous machine-based "sense-maker" develop research methods radically different from those we understand currently?

Could they specify novel sensors and observational techniques which we've not considered?

Ultimately, mankind might benefit from such an expansion in ways to understand the world around us, and understand the ways to understand the world - the philosophy of science.

Hi Paul,

I think the answers to almost all the points you raise have to be considered to be 'yes'. This makes the entire field of AI potentially dangerous unless it is properly regulated. Already we have deep learning algorithms connected to planet wide databases and capable of sifting through the date in accordance, we are assured, under solely human instruction. But for more advanced AI as you suggest, the instructions will be self formulated . . . and who knows what will happen then?Just what will these devices be learning to do? Don't forget that already deep learning AI beat the world best human GO player, and nobody knows how it did it, beyond iterating the largely banal 'it learnt how to do it'. So by the same token deep AIs will learn an awful lot more about all of us, and we will have neither any idea of what they have learned or what they intend to do with it.

Access to the IoT will empower deep AI to do almost anything humans can do and probably more unless formal restrictions are placed NOW on their capacities. That means developing metrics for AI, developing means of finding out exactly EVERYTHING that one has learned, and developing means of identifying with CERTAINTY what they intend to do with it.

Hi Tony

I agree with your concern and have reached similar conclusions as regards monitoring AI systems' progress and strengthening regulation where necessary, but hopefully without restricting "their" innovation and creativity, especially as regards how they perceive the world as they find it.

Of course, where they obtain access to our currently-expressed concerns, they will see that we think about "pulling the plug" and monitoring their progress - in response to which they might well start to behave as devious humans do - hiding their motivations and actions from regulative processes.

Perhaps we would need to treat them like "beloved children" and encourage them to be open and positive about biological beings too...

Otherwise, it becomes an exercise in Game Theory, playing against possibly sociopathic entities.

As I wrote over 40 years ago, in an early paper, perhaps we need an "ethical moderator" in such systems.

Regards - Paul

(I shall be returning to the question of how AI systems' interactions with the world might enhance mankind's understanding of how we understand the world, in a later post).