Wednesday 31 May 2023

Challenges for AI: if you're so bloody smart ...

It is suggested that the acid test for AI - indeed, for it to qualify as AGI - is that it should start generating New Science: hitherto unknown but recognisably scientific advances that can be tested against the usual criteria for such things.  We read that 'ordinary' 2023-vintage AI has been used in developing a wholly new antibiotic for use against Acinetobacter baumannii, a nasty strain of superbug.  OK, and all power to the team that achieved this: but it's not quite an AI achievement per se: we're still in "useful software" territory.  Couldn't have been done without software?  Well, neither could much of what I've been doing today by way of working from home.

So - I'd like to propose a test.  To my mind, one of the weirdest "non developments" of the past 100 years has been the complete failure of bio-chemical science to come up with artificial blood.  Surely, that would be up there with genuinely efficient large-scale storage of electricity** - crack that, and you can have your university re-named after yourself.  "Being a big help with finding antibiotics" is great.  Devising artificial blood - there's the real test.  Go for it, Mr A(G)I.

Any other tests that readers would like to suggest?



** Yes, of course I know that it's never really electricity that gets stored, it's electricity-into-something-else-&-back-into-electricity-again-later. 


dearieme said...

Drinks with the pleasant effects of alcohol without the unpleasant.

(To be fair I'm a fan of modern 0% beers but I've yet to see any kind comments on 0% wines.)

P.S. Are the pleasures of 0% beers partly accounted for by our previous experience of the pleasures of ordinary beers?

Anonymous said...

ND - why would any white Brit want to fight and die for people who hate him?

"I noted that the boards have recently been predominantly white male heavy, if we don't have enough BAME and female to board then we need to make the decision to pause boarding and seek more BAME and female from the RF [recruitment force].

"I don't really need to see loads of useless white male pilots, lets [sic] get a [sic] focussed as possible, I am more than happy to reduce boarding if needed to have a balanced BAME/female/Male board."

Another email sent the following day by the same squadron leader to a more senior member of the recruitment team, entitled "BOARDING PROFILE UPDATE", suggested a struggle to hit diversity targets.

It also discussed how efforts were being made to load more ethnic minority and female candidates on to selection boards.
To be fair, the Armed Forces are willing to defend the borders of Ukraine, or destroy the borders of Iraq, Libya or Syria, but I don't see the Navy patrolling off Dover, except perhaps as an expensive taxi service.

Anonymous said...

To be fair to the anon commander, I imagine his "useless" meant that he wouldn't be allowed to sign them up in any event, because the order from the top was for fewer white males.

Didn't there use to be laws against discrimination on the grounds of race? I'm sure I remember them...

andrew said...

My test for an AGI
Intelligent entities want things.
They want things that are not directly related to survival.
When they get what they want the reaction can be more complicated than a ticked box, they then want something else

djm said...

Any other tests that readers would like to suggest?

The evolution of a member of the political class that works to enrich the electorate as opposed to the current incumbents who evidently strive to exclusively look after No. 1

jim said...

I thought we had a number of potential blood substitutes - the snag is the cost. The deeper question is who needs and is prepared to pay for artificial blood - the usual stuff is common enough. As for batteries the key barrier is the link between the energy storing molecules and the electrodes that get electrons in and out. If we were able to couple the electrodes directly and intimately into the molecular storage structure we could do fast charging and rapid energy withdrawal. But that same new battery potentially becomes a bomb. Be careful what you wish for.

So the blood problem is mainly economics and the battery problem one of new ability we don't have. Neither is solvable by trawling a huge and fancy database.

The key missing piece in AGI is a way of joining up all the implications of the little bits of data we do know and spotting the dark corners we don't know. For example Maxwell's laws for radio and light waves came from realising there was something missing from the Victorian mathematical equations for electricity and magnetism. The missing link was not in the database. For Einstein and the speed of light all the pieces of data were there, all that was necessary was some peace and quiet and clear thinking. The link was in the database but no one had spotted it. Einstein and gravity, all the pieces were there but needed a bit of clear thinking and some hard sums.

The key problem with AGI is that we have a fancy database hooked up to a very limited implications engine. But we are relying on the notion that if we put enough NVIDIA chips together some kind of intelligence will emerge even if we don't know why. No one has found anything in the grey slop in our heads that looks like 'intelligence' so more of the same is the only way to go - for now.

Meanwhile the physical sciences have been a bit moribund these last 20 years. No time machines etc. Maybe AGI will help find a few useful breakthroughs - or not.

dearieme said...

Presumably there will be people prepared to pony up for blood transfusions free of spike protein. Work there for AI?

BlokeInBrum said...

To anyone interested in machine learning/artificial intelligence and other related topics I highly recommend "Gödel, Escher, Bach: an Eternal Golden Braid" from Douglas Hofstadter.

Don Cox said...

Spike protein disintegrates after a few hours, so there would be no need for special spike-free blood.

Don Cox

dearieme said...

"Spike protein disintegrates after a few hours"; I dare say - but whose word are you taking on that? Dr Fauscist?

dearieme said...

@Don, the internet reports "The Infectious Disease Society of America (IDSA) estimates that the spike proteins that were generated by COVID-19 vaccines last up to a few weeks, like other proteins made by the body."

On what basis it forms that conclusion I don't know. But if Covid taught us nothing else, we know we can't trust the medical trades.