The game of Go has long been considered one of the most complex and challenging games for artificial intelligence (AI) to master. However, in recent years, AI has made significant progress in this area, with machines like AlphaGo and AlphaGo Zero beating even the best human players.
That’s why it was so surprising when a team of human players, with the help of a bot, recently defeated the top AI program in the world at a Go tournament. This victory has reignited the debate about the limits of AI and what it means for the future of human-machine collaboration.
An amateur Go player competed against a highly-ranked AI system and, after exploiting a flaw identified by another computer, was successful in gaining the victory, reported The Financial Times.
Kellin Pelrine’s recent exploit of a serious flaw rendered in KataGo demonstrated the power of human masterminds. This monumental victory — 14 out of 15 games with no computer assistance since AlphaGo’s trailblazing advance in 2016 — shows that even highly advanced AI can have considerable shortcomings.
Pelrine was able to secure a victory thanks to the efforts of the research firm FAR AI, who employed their self-developed program, KataGo. After over a million games tested against this program, Pelrine found flaws enabling them to win as amateur players.
He used the same method to defeat Leela Zero, another top Go AI, as he had used previously.
“Not completely trivial but it’s not super-difficult.”
This strategy aims to create a “loop” of stones around the opponent’s group while simultaneously making moves elsewhere on the board, thus diverting their attention. Despite almost complete encirclement, the computer still did not detect the tactic being employed.
Pelrine went on to say:
“As a human, it would be quite easy to spot.”
Said he, the stones on the board encircle are so obvious that they can easily be identified.
AI systems can be too limited in their thought processes to pass human standards, showing off clear stupidity in many instances. This proves that AI narrows its speculative capacities to essentially being what it is taught.
Microsoft’s Bing search engine chatbot provides a relevant example of the drawbacks that AI systems face even when doing basic tasks. It successfully managed tedious tasks such as devising travel plans but perplexed users with erroneous information and sometimes exhibited outrageous behavior likely due to its learning experiences.
Lightvector, the developer of KataGo, having been made aware of problems being exploited by players over several months, has stated their progress in work to fix attack types that have taken advantage of this exploit on their GitHub post.
The recent victory of a team of human players, aided by a bot, over the top AI program in the world at a Go tournament is a remarkable accomplishment that highlights the potential for humans and machines to work together to achieve even greater success. While AI has made impressive progress in recent years, it’s clear that there are still areas where humans excel, such as intuition, creativity, and adaptability.
As we continue developing and refining AI, we must recognize humans’ and machines’ unique strengths and limitations. By combining our complementary skills, we can solve complex problems and achieve outcomes that would be impossible with either one alone.