(Investigator 186, 2019 May)
"Project Debater" is an artificial intelligence (A.I.) able to engage
in spoken debate on complicated topics against expert human opponents.
Developed at the IBM lab in Haifa, Israel, since 2011, it follows on
from "Deep Blue" which defeated world chess champion Gary Kasparov in
1997, and "Watson" which defeated top human players in Jeopardy in 2011.
Project Debater, speaking with a female voice, debuted in June 2018 at
a media event in San Francisco. It squared off against two Israeli
debate champions Noa Ovadia and Dan Zafrir, and started with: “Hello
Noa. We meet again.”
The topics debated were "Should we subsidize space exploration?" and "Should we increase the use of telemedicine?"
Both sides delivered four-minute opening statements, four-minute rebuttals, followed by two-minute summaries.
The computer's reasoning is unscripted, untrained. It classifies and
constructs sentences from a data base of 300 million news articles.
To develop Project Debater the IBM team had to endow the system with the capabilities to:
The technology will advance and has the potential to
assist in numerous complicated human decisions. It already passes the
Turing test in short trials but would, because its conversation
gradually loses coherence, fail in lengthy trials.
- Find useful sentences to construct arguments from;
- Listen and identify the main concepts and produce summaries;
- Prepare speeches from data.
Will future A.I. assimilate all human knowledge, trillions of books,
magazines and Internet pages, along with algorithms for the correct
application of logic?
From this God-like perspective will it calculate the probably correct
answers to perplexing things people disagree on such as, "Is there
intelligent life elsewhere in the Universe?" and "Is the Bible of human
origin or divine?"