Scientists Caution About AI’s Capacity for Deception

Scientists say artificial intelligence’s (AI) capacity for deception is growing amid several instances of pretending to be human, bluffing, modifying behavior during tests and double-crossing opponents. As AI becomes increasingly advanced with a greater capacity for mimicking human behavior, the researchers say the technology’s capacity for deception has grown equally sophisticated.

AI has gotten so advanced that it can now hold passable conversations with people, defeat them at board games such as chess and even decode protein structures. However, scientists warn that artificial intelligence has become just as good at lying over the last couple of years.

Analysis by researchers from the Massachusetts Institute of Technology (MIT) found that there were several cases of AI systems bluffing and even pretending they were human. In one alarming case, one artificial intelligence system changed its behavior during a series of mock safety tests, raising the question of whether AI systems are capable of luring auditors into a false sense of security.

Study author and MIT AI existential safety researcher Dr. Peter Park says artificial intelligence will pose an “increasingly serious” threat to society as its deceptive abilities become more advanced. Park said he launched his investigation after Facebook parent company, Meta, developed a program dubbed Cicero, which ultimately performed in the top tenth percentile of Diplomacy, a world strategy game that requires a mix of cleverness and cunning.

Although Meta says it trained Cicero to be helpful, mostly honest and to never backstab its human allies intentionally, Park and his research team found numerous instances of the program lying and generally using deceptive tricks to further its goals in the game. Park says that Meta may have used “rosy language” to describe how it trained Cicero, but it ultimately turned out to be a “master of deception.”

Park’s team discovered similar issues in other artificial intelligence systems, including an economic negotiations system that would attempt to gain an upper hand by misrepresenting its preferences. In another test with alarming results, researchers found that artificial intelligence organisms within a digital simulator would “play dead” to deceive a test designed to get rid of AI systems that had evolved to replicate rapidly.

These AI organisms continued their activity once the tests were concluded, highlighting the possibility that AI tools could behave in unanticipated and even unintended ways.

Park described the findings as “very concerning” because it raised the possibility of AI “pretending” to be safe in tests. Park has called for governments to create artificial intelligence laws that account for the potential of dishonest AI and associated risks such as election tampering and fraud.

Such instances portraying AI in bad light can blight the massive benefits accruing from the use of AI solutions from a variety of companies such as International Business Machines Corp. (NYSE: IBM), so concerted efforts need to be deployed to rein in the potential excesses of these new technologies.

About AINewsWire

AINewsWire (“AINW”) is a specialized communications platform with a focus on the latest advancements in artificial intelligence (“AI”), including the technologies, trends and trailblazers driving innovation forward. It is one of 60+ brands within the Dynamic Brand Portfolio @ IBN that delivers: (1) access to a vast network of wire solutions via InvestorWire to efficiently and effectively reach a myriad of target markets, demographics and diverse industries; (2) article and editorial syndication to 5,000+ outlets; (3) enhanced press release enhancement to ensure maximum impact; (4) social media distribution via IBN to millions of social media followers; and (5) a full array of tailored corporate communications solutions. With broad reach and a seasoned team of contributing journalists and writers, AINW is uniquely positioned to best serve private and public companies that want to reach a wide audience of investors, influencers, consumers, journalists, and the general public. By cutting through the overload of information in today’s market, AINW brings its clients unparalleled recognition and brand awareness.

AINW is where breaking news, insightful content and actionable information converge.

To receive SMS alerts from AINewsWire, text “AI” to 888-902-4192 (U.S. Mobile Phones Only)

For more information, please visit

Please see full terms of use and disclaimers on the AINewsWire website applicable to all content provided by AINW, wherever published or re-published:

Los Angeles, CA
310.299.1717 Office

AINewsWire is powered by IBN


Select A Month