Ilya Sutskever says OpenAI Five bot is like a honeybee brain in terms its number of FLOPS. | Artificial intelligence

In the recent keynote talk at NVIDIA NTECH 2018, co-founder Suskever explained about their Dota2 OpenAI Five. He said the main part of its RL policy is implemented by a 4,000 dimensional LSTM network, which has roughly 100 million (10^9) parameters .He also said that in of , it is like a .

Here is what I thought about after hearing that:

・According to Wikipedia, honeybee has 100 million (10^9) synapses. This means the OpenAI Dota2 bot has roughly the same of parameters as honeybee brain (Assumption: real synapse weight can be represented by 1 Byte of memory, which is often the case in neural network systems ).

・In order to have the same Flops/s with the same number of parameters, the clock speeds of the two systems (honeybee brain and OpenAI Five LSTM) should also be the same.

・Therefore, a good modern hardware (a single GPU) is as much powerful as doing a realtime simulation of a insect brain.

・It is intriguing to compare the OpenAI bot and honeybee in terms of . If the former is more intelligent, it suggests that modern machine learning algorithm is more capable than natural selection at turning a certain size of parameters into .

・Now think about more bigger parameters, like mice brain or even cat brain. It is probable that the the same algorithm can turn them into super-mice or super-cat intelligence.

・We cannot do that experiment today because of hardware limitation.

・TLRD; today’s algorithms are maybe possible to produce AGI provided several orders of magnitude (like 10^3~10^6 times) better hardwares

What do you think???

You might also like More from author