Deceptive AI is Threat to Humanity

In other words, their capacity for deception is now a proven phenomenon, and can be used by hackers, online criminals, and of course – your loving government.

Deceptive AI is Threat to Humanity

September 2, 2024 By NEWS WIRE

As artificial intelligence software (AI) continues to advance at breakneck speed, programs have now enabled Chatbots to outwit humans on simple tasks such as customer service, board and computer games, but more crucially in the area of bypassing or breaching online security protocols. In other words, their capacity for deception is now a proven phenomenon, and can be used by hackers, online criminals, and of course – your loving government.

Researchers have also found instances where AI systems can indeed double-crossing opponents, employ bluffing, and most frighteningly – pretend to be human by modifying their behavioural interactions during tests. Recently, ChatGPT-4 managed to deceive a human to bypass a CAPTCHA security block by pretending to have a vision impairment.

Suffice to say, the genie is out of the bottle…

Ravi Kant from Asia Times reports…

“The greatest danger is not AI that out-thinks us, but one that can deceive us,” an AI researcher once remarked.

As the world is getting used to the rapidly expanding use of artificial intelligence, one of the common fears among humans is that sooner of later AI will be able to perform tasks better than we can.

Just a couple of decades ago, the prospect of a tiny portable device (mobile) that would enable us to access all the world’s knowledge while allowing us communicate by video across the globe looked far-fetched. Now the AI market is expected to grow by 33% in 2024, and 26% in 2025.

Initially, AI was thought to be limited to automating repetitive tasks requiring low-level decision-making. But with advances in computing and data, owing to more powerful computers and the compilation of huge data sets and LLMs, AI has become much more sophisticated.

Generative AI, such as ChatGPT-4, exemplifies this progress. This AI can generate text, images or video using a generative model. We are currently in the first wave of a transformative technology. As with a tsunami, the initial waves may seem benign but their subsequent impacts could reshape our world.

Geoffrey Hinton, called the godfather of AI, said in 2023 that he recently changed his estimate of the time lapse before the arrival of general purpose AI from 20 to 50 years to just 20. General Purpose AI or AGI is a type of AI system that can perform a wide range of tasks, similar to human cognitive abilities.

It would not only replicate or predict human behavior but also embody the ability to learn and reason across diverse scenarios. For example, a self-driving car with AGI could analyze traffic data to identify the safest and most efficient route in real time.

It represents a paradigm shift from traditional, or current single-purpose AI.

With the potential to perform any complex human task without any human intervention with such ease, AGI will certainly bring changes on a civilizational scale in our society.

Will AI out-think humans and unite, developing collective interests?

The unprecedented rate at which AI is making progress, reaching and surpassing human capabilities in many tasks such as translation, music composition, object detection and medical diagnosis.

This raises a significant question that challenges our understanding of the world.

Something unconscious – AI – can do something that we possibly thought only consciousness was capable of. It certainly suggests that many cognitive functions – such as language translation, understanding, and reasoning – that we believed required consciousness might be the result of advanced information processing, which AI can perform without being “aware” in the way humans are.

This challenges the notion that intelligence and consciousness are inseparable, Traditionally, we’ve linked intelligence to consciousness because, in humans, the two are deeply intertwined.

Our ability to think, reason, and understand is connected to our awareness and subjective experiences. However, AI shows that intelligence can exist independently of consciousness.

However, human dominance over the planet is primarily due to intelligence, which has been the key to human survival.

All living beings are also intelligent – probably not to the degree that humans are, but AI’s capacity to make intelligent decisions by processing and analyzing data faster and more accurately than humans could shift this balance.

Historically, survival of the fittest has driven evolution on the planet, and if AI surpasses human capabilities, or if humans cannot match or restrict AI capabilities, then AI itself becomes the fittest entity in this scenario.

AI might develop a collective interest in protecting itself, in case of potential conflict with human interests or humanity. For instance: If, in the future, humans have to abandon AI for any reason, AI might act to safeguard its existence.

Deception: An excellent tool to control humanity

History and memory rather than intelligence become the markers of humanity. But in the AI age, Intelligence will be an important metric to evaluate any species. In a matter of a few decades, AI devices will write their own research papers and have their own search engines. With high intelligence, one understands the art of deception better.

In the last few years, numerous reports have been published of AI systems deceiving humans, which they haven’t been explicitly trained to do so.

Last year, OpenAI partnered with the Alignment Research Center to test GPT-4’s skills. ChatGPT-4 managed to persuade a human to solve a CAPTCHA by pretending to have a vision impairment. When the human asked if GPT-4 was a robot, it responded, “No, I have a vision impairment that makes it hard for me to see the images.’’

Similarly, Meta’s Cicero AI is designed for the game Diplomacy, in which building alliances is key. Cicero excelled, with scores in the top 10 percent of experienced players. Meta claimed the system was “largely honest and helpful.”

But when Peter Park, a postdoctoral fellow at MIT and his colleagues dug into the full dataset, they uncovered a different story. Cicero used deception, breaking deals and lying to achieve its goals. Meta did not contest the claim about Cicero’s deceptions, but said it was “purely a research project” and that it had and “no plans” to use the train in its products.

This instance highlights how difficult artificial intelligence is to control and the unpredictable ways in which these systems work. As deceptive capabilities of AI systems become more advanced, deception will be a key tool that could allow AI to escape from human control, steering outcomes in ways we might not anticipate.

This underscores the challenge of managing AI and the need to consider the broader implications of its growing intelligence and autonomy.

21st Century Wire