Is a Roomba as smart as a housefly? (controller, robot, format)

[Avondalist]

A fly got stuck in my house today because I had the door open. It kept trying to fly through the sliding screen door to get back out. I noticed and opened the screen door halfway. I tried to shoo the fly out for several minutes, but it just kept landing back on the screened-in portion where it had last been. Finally by some chance it flew out of the open portion.

That got me thinking about Roombas, how they vacuum the same part of the floor over and over and miss other parts. A lot of artificial intelligence is pretty primitive now, but so is a lot of biological intelligence. I wondered if you could start comparing various AI systems to organisms.

I found this pop-science article saying no, none of the artificial intelligences we have are comparable to any animal. https://bigthink.com/mike-colagrossi...y-smarter-than

But then I've also seen studies where AI can achieve an above-average test result on an IQ test. AI is capable of besting biological intelligence at certain tasks, while biological intelligence is capable of besting biological intelligence at other tasks.

In terms of motor control and mapping out an environment to maneuver in, I think the housefly has the Roomba beat. The Roomba might have the housefly beat when making decisions, but that would be hard to know because the housefly and the Roomba brains are trying to solve different problems.

Even a smart AI system like GPT-3 might be considered dumber than a housefly, because it requires so much more energy and memory to complete its tasks than a biological brain. I have no doubt that per unit of space taken and per unit of energy consumed, biological intelligence is more developed than artificial intelligence. However when AI can scale up it can outperform certain tasks.

It can also learn faster. GPT-3 was released a little more than a year after GPT-2, so the actual training took probably much less than that. Humans are not able to learn how to reply to questions or translate between languages in less than a year; it takes us years to learn how to speak and years more to develop conceptual models and vocabulary to do what GPT-3 does.

Along some dimensions AI outperforms, and along others it lags. I'd be interested to see if there have been any more exhaustive looks at comparing species to AI than that one article I found.

[Harry Diogenes]

I develop AI systems that use a small number of artificial neurons, but they do dedicated tasks. For a game controller, I use 3 neurons and can train an agent in a game to run, hide, or attack based on the state of itself and it's enemies (health, armed, number of enemies, etc).

But to create a robot that could do that and actually move on legs would take much more processing, and I doubt my serial processing systems would be quick enough for a robot.

This is the same for the Roomba, which is dedicated to do one task, but I could hit it all day with a rolled up newspaper, something I could not do with a fly. I could also train a system to find a path round a half open screen door that would outsmart a fly, but the fly would outsmart my system in everything else.

A fly's brain is responsible for hunting food, avoiding enemies, flight, detection of air currents, etc. And a fly's brain processes in parallel, just as our brains do, and uses rules of thumb processing as we do, both which allows a fly to react quicker.

So perhaps a better comparison could be to see which is better at processing information than to see which is more intelligent.

[mshultz]

If a pet defecates on the floor, both the fly and the Roomba will be attracted right to it, but the Roomba will make a much bigger mess.