Turing Lecture: Beneficial AI for the advancement of humankind

I attended the annual Turing Lecture on Wednesday 22nd February 2017, on the topic of "AI of the advancement of humankind" with Dr Guruth Banavar of IBM Cognitive Science giving the talk. Firstly, he's been involved a long long time, since before the AI Gammon (IBM's first machine built to take on, and beat anyone at Backgammon) days it would appear. And he's still involved in the latest ongoing work including the shining light in IBM's stable at present, Watson.

My raw notes from the event can be found here

Update 03/03/17: the full, official, recording of the talk can be found here.

And if you'd rather hear a compressed version of the notes, and in audio, here's my first podcast. Firstly, it's unedited so the start and end is a little rough, apologies. But the important bit where i discuss the event is good. All thoughts welcome.

 

My initial notes here outside of just what I heard was that it was an excellent talk, routed in strong facts (albeit with a bit of IBM spin which you can only expect) and grounded ideas of where we are and Where We Are Going (sign-up form for my fortnightly mails on the latest news and why). The killer piece was the end when a question to the audience was put to him on whether or not, Watson's achievements at the game of Jeapordy was truly 'intelligence' or was it 'brute force' (i.e. Watson essentially scours every available source of data, and through some ranking tricks, picks the most probable answer). Dr Banavar was honest enough to straight out say that the answer given realistically was just a brute force approach - but also honest enough to recognise that the betting aspect of Watson in the game was doing some interesting ideas. When someone is open enough about their own product to recognise its failings, it gives me even more of an appreciation of them.

 

My own thoughts on AI at present - I've written on this elsewhere - is that right now, much of AI is purely just making use of the incredible performance of modern computing and cloud computing systems to do what was, as described above, is a heavy amount of machine learning and brute force. I.e. it's not 'intelligence' in the strictest sense, but it is a hugely powerful use of computing machines for what they are best at: working on huge quantities of data to see trends and routines as well as making the most of their huge capability to assess thousands, or millions, or possibilities to see the most optimum outcome. Hype around a truly smart machine, smarter than humans is still a long way away right now - and I see that most of the smart people agree, and call this Artificial General Intelligence (AGI), thus separating from the current hype around AI.

I'll elaborate more on this next week in another post but wanted to get my notes out straight off the bat while fresh. Would love to hear more from others......

Neal McQuaid