So much for AI being able to beat players in DOTA

Vastet
atheistBloggerHigh Level ModeratorSuperfan
Vastet's picture
Posts: 13131
Joined: 2006-12-25
User is offlineOffline
So much for AI being able to beat players in DOTA

https://www.pcgamer.com/open-ai-bots-fall-to-human-dota-2-players-at-the-international-2018/

AI is so far from being a threat it's really ridiculous how some otherwise intelligent people freak out about it.

Proud Canadian, Enlightened Atheist, Gaming God.


Beyond Saving
atheist
Beyond Saving's picture
Posts: 5420
Joined: 2007-10-12
User is offlineOffline
 Hardly a fair fight, the

 Hardly a fair fight, the AI is an infant while the humans have dedicated their lives to playing the game (or similar games) for years. A true test would require the AI to play tens of thousands of games against a wide variety of opponents and then see what it learns. It sounds to me that the computer didn't have a firm grasp on the rules (it didn't understand that maximizing gold was important). Kind of like early iterations of chess computers failed to account for the end-game structure. Now, modern computers can easily trounce the best players in the world. 

If, if a white man puts his arm around me voluntarily, that's brotherhood. But if you - if you hold a gun on him and make him embrace me and pretend to be friendly or brotherly toward me, then that's not brotherhood, that's hypocrisy.- Malcolm X


Vastet
atheistBloggerHigh Level ModeratorSuperfan
Vastet's picture
Posts: 13131
Joined: 2006-12-25
User is offlineOffline
I might agree if not for the

I might agree if not for the fact that the exact same circumstances were in play a year or two ago and the AI won the game, and that victory was used to propagandize the threat of AI. Especially as another AI had just come off winning a Go game or series against a champion Go player. The AI also has an insurmountable reaction time which can easily make up for shortcomings in other areas. Something that takes a human half a second takes a computer 1/100th a second or less.

What AI will never grasp is strategy. That is why gold per minute was noticeably deficient on the AI side. In MOBA games, strategy is very important. Gold income is dependent on strategy.

MOBA's are simple enough that I think eventually a strategy of sorts can be put in the code, but the more complex a game is the less that will be possible. If you completely opened up the game so all characters could be used, even 10's of thousands of games wouldn't help the AI.

Proud Canadian, Enlightened Atheist, Gaming God.


EXC
atheist
EXC's picture
Posts: 3804
Joined: 2008-01-17
User is offlineOffline
Beyond Saving wrote: Kind

Beyond Saving wrote:

 Kind of like early iterations of chess computers failed to account for the end-game structure. Now, modern computers can easily trounce the best players in the world. 

If you read the accounts of when a computer first beat the world chess champion, Garry Kasparov gave up due to fatigue, not because Deep Blue was more intellegent. If the human chess champion could take a month to decide each move they would win(assuming a computer still had minutes to decide and could not be parallel), but the games may take 10,000 years. To win at DOTA, they just need to use more parallel compute resourses. As long as there is time limit for the humans to make decisions, computers will win.

The programmer that wrote the "AI" code could work out all the calulations on paper and perform whatever operation an AI computer does if given enough time. Computers aren't more intelligent just faster. AI is another scam until proven otherwise. Even human intelligence(e.g. IQ tests) are largely a scam. Such is life in a highly competitive world.

 

 

 

Taxation is the price we pay for failing to build a civilized society. The higher the tax level, the greater the failure. A centrally planned totalitarian state represents a complete defeat for the civilized world, while a totally voluntary society represents its ultimate success. --Mark Skousen