There seems to be a bit of emperor’s new clothes going on, at least in fintech: I know it’s not AI, you know it’s not AI, now let’s just carry on. But it’s not just related to AI. For pretty much any buzzword in technology funding, I’ve been in a situation where a supplier says to a customer “we do $buzzword” and they do their due diligence and exclaim that wow, this one really is doing that.*
The reason that fintechs generally don’t have AI in their core business is that they don’t have the data. If you’re new and small, you don’t have access to the data. If you’re big and old, then a lot of it is still hand-written on paper and not amenable to processing.
All except blockchain. The blockchain companies all seem to run their own internal database then write a receipt of a transaction to a public ledger.
Define AI. I don’t consider machine learning to meaningfully be AI. Machine learning is surprisingly useful for a variety of tasks and research will likely push it in new directions.
AI in this context is a marketing term, and kudos to anyone who can sell on it without misleading their customers about results and limitations.
Define AI. I don’t consider machine learning to meaningfully be AI.
This is the path that leads to humpty-dumptyism: AI means precisely what I mean it to mean, no more and no less. The “AI effect” is where anything that can be done by today’s computers is not AI, and discounting machine learning because of the AI effect is to ignore two strong claims in favour of it being AI:
artificial neural networks (admittedly only one class of ML system) were a deliberate attempt to model existing intelligent systems in researching artificial intelligence;
machine learning agents are capable of taking information from their environment and using it to aid in achieving their goals, a textbook definition of AI (adapted from Norvig and Russel 2003).
Not everything that comes out of AI research is AI (the hashmap, for example, is not); not everything that comes out of it is not AI.
Your point 2 sneaks in a couple of things that are in fact not present in most applications of machine learning: an agent interacting with an environment and learning from its environment.
This doesn’t describe pretty much any deployed system. Agents used for control tasks would be, but let’s be honest: self driving vehicles are still a way off, and the other major class of such agents seems to be players of games.
Finally, I don’t consider agents that don’t learn from real experience (as opposed to just training runs) to be AI. The ability to learn from new experience is a key part of intelligence; loss of that ability is a serious illness for humans.
Well hold on, serious illness and object are very different things. If we’re going to argue the loss of ability to learn means we can dispose of them as machines, that mostly just shows how learning is not strictly speaking fundamental to intelligence.
100% of “AI startups” don’t use AI, since AI hasn’t been invented yet. The best we have is machine learning, which can be useful, but is only a very distant approximation of AI in the best of cases.
Many people seem to mean different things when they talk about “AI”, and I’m fairly sure there is a lot of confusion among people what “AI” really is, especially slightly less technical people (and even among some technical people who haven’t looked in to the matter).
I’m not massively purist on these matters, but claiming that ML is AI is misleading marketing, at best.
I don’t think it’s marketing. I think actually ML as a term is marketing. AI techniques are merely techniques that are often inspired by biology and attempt to mimic intelligence. The science fiction associations with the term are merely science fiction. If someone has some fantastical view about what computers are capable of the word you use to describe it isn’t going to help much. There is no bar to cross for something be “AI” it is a gradient of proficiency.
There seems to be a bit of emperor’s new clothes going on, at least in fintech: I know it’s not AI, you know it’s not AI, now let’s just carry on. But it’s not just related to AI. For pretty much any buzzword in technology funding, I’ve been in a situation where a supplier says to a customer “we do $buzzword” and they do their due diligence and exclaim that wow, this one really is doing that.*
The reason that fintechs generally don’t have AI in their core business is that they don’t have the data. If you’re new and small, you don’t have access to the data. If you’re big and old, then a lot of it is still hand-written on paper and not amenable to processing.
I thought it would be 100%. Downgrading the meaning of “AI” to mean “machine learning” isn’t fine.
Define AI. I don’t consider machine learning to meaningfully be AI. Machine learning is surprisingly useful for a variety of tasks and research will likely push it in new directions.
AI in this context is a marketing term, and kudos to anyone who can sell on it without misleading their customers about results and limitations.
This is the path that leads to humpty-dumptyism: AI means precisely what I mean it to mean, no more and no less. The “AI effect” is where anything that can be done by today’s computers is not AI, and discounting machine learning because of the AI effect is to ignore two strong claims in favour of it being AI:
artificial neural networks (admittedly only one class of ML system) were a deliberate attempt to model existing intelligent systems in researching artificial intelligence;
machine learning agents are capable of taking information from their environment and using it to aid in achieving their goals, a textbook definition of AI (adapted from Norvig and Russel 2003).
Not everything that comes out of AI research is AI (the hashmap, for example, is not); not everything that comes out of it is not AI.
Your point 2 sneaks in a couple of things that are in fact not present in most applications of machine learning: an agent interacting with an environment and learning from its environment.
This doesn’t describe pretty much any deployed system. Agents used for control tasks would be, but let’s be honest: self driving vehicles are still a way off, and the other major class of such agents seems to be players of games.
Finally, I don’t consider agents that don’t learn from real experience (as opposed to just training runs) to be AI. The ability to learn from new experience is a key part of intelligence; loss of that ability is a serious illness for humans.
Well hold on, serious illness and object are very different things. If we’re going to argue the loss of ability to learn means we can dispose of them as machines, that mostly just shows how learning is not strictly speaking fundamental to intelligence.
100% of “AI startups” don’t use AI, since AI hasn’t been invented yet. The best we have is machine learning, which can be useful, but is only a very distant approximation of AI in the best of cases.
It’s kind of true that people talk of:
When people say “AI” the do not mean a general AI, and never have. Why would you think they mean this now?
Many people seem to mean different things when they talk about “AI”, and I’m fairly sure there is a lot of confusion among people what “AI” really is, especially slightly less technical people (and even among some technical people who haven’t looked in to the matter).
I’m not massively purist on these matters, but claiming that ML is AI is misleading marketing, at best.
I don’t think it’s marketing. I think actually ML as a term is marketing. AI techniques are merely techniques that are often inspired by biology and attempt to mimic intelligence. The science fiction associations with the term are merely science fiction. If someone has some fantastical view about what computers are capable of the word you use to describe it isn’t going to help much. There is no bar to cross for something be “AI” it is a gradient of proficiency.