"General Artificial Intelligence" Might Be Impossible, But It's Definitely Unprofitable

Posted by Alyssa Verzino on Apr 24, 2018 1:15:00 PM

face trace-3157431_1920

Everyone from Elon Musk to the late Stephen Hawking to an entire research institute are warning about the dangers of an A.I. apocalypse, where some real-life version of SkyNet or HAL 9000 will rise up to wipe out pesky human life. This is probably alarmist, not just because general artificial intelligence like we see in the movies may be impossible, but because -- for the foreseeable future -- there's no money in it.

And if there's no money in the apocalypse, humans generally don't rush to bring it about.

General artificial intelligence -- also known as "strong A.I." or "human-like A.I." -- is what we find in science fiction. These are software programs (often inhabiting robot bodies) that can reason and learn independently in the same fashion as the human mind, but often at the speed and scale of a massive supercomputer. We're nowhere near this technology, not least because we don't really understand how it is that human brains can think in abstract terms and generalize ideas from one concept to another.

It's a hard problem, wrapped up in neuroscience, biophysics, and even philosophy. And it's such a hard problem, it will likely take billions upon billions of dollars and years upon years of research to crack, assuming it ever really becomes possible.

In the meantime, there are easier ways to make money on A.I. right now, and so those billions of needed dollars are unlikely to be pointed at generalist A.I. research anytime soon. Specialist A.I. is where the money is at.

Specialist A.I. is a form of artificial intelligence that is designed and trained to perform one narrow task as well or better than humans. For example, recognizing faces. Or, perhaps more usefully, recognizing parked firetrucks.

In the near future, these hyper-specialized A.I. skills will be strung together to perform very specific jobs instead of humans. For facial recognition, it may replace the work of a doorman (or just tagging photos on Facebook so you don't have to). For firetruck recognition, it will be part of a larger suite of self-driving car skills that take humans out of the driver's seat.

But it's important to understand that the A.I. recognizing faces doesn't understand that those faces belong to people, and we can't apply that same basic facial recognition algorithm to firetruck recognition or any other visual task. That requires training up a whole new algorithm. There is no deeper learning here; it's just sets of skills made better and better all stuck together to accomplish more complex tasks.

The more solvable problem -- and the more useful software -- is one that, for example, can parse out requests for tech support from a slack conversation, or recognize human resources requests inside an email. String that together with an A.I.-friendly knowledge base and chatbot, and you can start replacing Tier 1 tech support or an HR intern.

Go a step further and you can have the A.I. manage the knowledge base so that the chatbot never lacks for an answer, and pretty soon you've got your documentation staff automated, too. Combine that principle with a blockchain ledger to let multiple companies combine and remix their specialized A.I. agents, and you've got a universe of useful artificial intelligence that will never even come close to assassinating astronauts or building Terminators.

If you'd like to talk about how specialized A.I. can prepare you for the real future, not the scary TV version, contact Talla today.

Topics: AI, Artificial Intelligence, future of ai