Will we be smarter than humans in 5 years? The breakneck pace of AI

Jeffrey Hinton, one of the “godfathers of AI,” has been particularly outspoken since leaving Google earlier this year. He is credited with perfecting and popularizing backpropagation, a pivotal algorithm that allows multilayer neural networks to correct mistakes.

This breakthrough contributed to the success of deep learning technology, which is the backbone of today’s generative AI models. For his groundbreaking contributions, Hinton received the Turing Award, considered the Nobel Prize of computer science.

pace of progress

Hinton went from being an AI optimist to being an AI pessimist. noticed He said AI will probably be smarter than humans within five years, rather than the 50 to 60 years he thought. Last spring, he warned of the potential existential threat posed by AI that could soon be smarter than humans. The reason for his growing concern is the huge leap forward we are seeing in generational AI through large-scale language models (LLMs).

Five years from now is 2028, and that prediction is even more aggressive than that of AI optimist and Google head of engineering Ray Kurzweil.

“By 2029, computers will have human-level intelligence.” Kurzweil said In an interview a few years ago. He further predicted By 2045, AI is thought to reach the “singularity,” a point where “effective intelligence increases a billion times by merging with the intelligence we create.”

in the last 60 minutes interview, Hinton argued that current major AI models, such as those developed by OpenAI and Google, already have real intelligence and reasoning capabilities. Remarkably, these models can have unique sensory experiences similar to humans, he added. Hinton said that while he does not currently believe that AI systems are conscious (in the general sense of the concept), over time they will become conscious.

AI growth stages

Hinton believes that in five years, advanced AI models “will likely be able to make better inferences than humans.” When asked if humans would become the second most intelligent species on Earth, Hinton answered, “Yes.” He further added: “I think my main message is there’s a lot of uncertainty about what’s going to happen.” [going to] What happens next? I understand these things. ”

We seem to be entering a phase of AI growth. Just like when parents need to be careful what they say in front of their children. “And they understand, so we need to think hard about what happens next,” Hinton added.

With the acceleration of development only increasing, it is clear that we need to act now. Recent developments have put to rest any doubts about whether an AI arms race is underway. Specifically, CNBC report China plans to increase its computing power by 50% by 2025 as it seeks to keep pace with the United States in AI and supercomputing applications. This is massive computing power to build and train LLMs on a larger scale than ever before.

Next generation LLM

according to According to Hinton, the human brain has approximately 100 trillion neural connections. In contrast, today’s largest AI systems have only 1 trillion parameters. However, he believes that the knowledge encoded in these parameters is far beyond human capabilities. This suggests that AI model learning, especially knowledge retention, is much more efficient than human learning.

in addition, report The next generation of LLM is coming soon, perhaps by the end of this year, and could be 5 to 20 times more advanced than the GPT-4 models currently on the market.

Mustafa Suleiman, CEO and co-founder of Inflection AI and co-founder of DeepMind, predicts in The Economist conversation “In the next five years, frontier model companies (leading companies training very large AI models) will be training models that are more than 1,000 times larger than what we see in GPT today. 4. ”

These larger models have great upside potential. These tools not only serve as highly capable personal assistants, they also help us solve our greatest challenges, such as nuclear fusion reactions that create limitless energy and provide precision medicine for longer, healthier lives. may help solve the problem.

The concern is that as AI becomes smarter than humans and develops consciousness, its interests may diverge from those of humanity.

Will it happen? If so, when will it happen? Hinton says: “We don’t know that.”

Governance challenges

Technological advances in AI are impressive, but they are also putting significant pressure on global governance. Another AI race — Government regulation of AI tools. However, the speed of AI development places significant burdens on regulators. They need to understand technology and how to regulate it without stifling innovation.

The EU is facing these issues and is thought to be nearing the final stage of discussions. comprehensive law (AI Law).However, recent report The US has indicated that it believes the EU law will hurt small and medium-sized businesses while favoring companies with the financial resources to cover the costs of compliance, “suppressing expected productivity gains.”

This concern at least suggests that the United States may pursue a different approach to regulation. However, regulations from other countries could fragment the global landscape of AI regulation. This reality can potentially pose challenges for companies operating in multiple countries, as they must comply with and navigate different regulatory frameworks.

Additionally, this fragmentation can stifle innovation if small businesses cannot bear the costs of compliance in different regions.

Time to change?

However, there may still be potential for global cooperation in AI regulation.according to register, G7 leaders are expected to enact international AI regulations by the end of the year. At the G7 earlier this year, agreed Establish a Gen AI-related working group to discuss governance, intellectual property rights, disinformation, and responsible use. However, China is conspicuously absent from this list of countries, along with 24 other EU member states, raising questions about the impact of the G7 agreement.

In an interview with 60 Minutes, Hinton also said: [when] We look back and see that this was a kind of tipping point where humanity had to make decisions about whether or not to develop these things further and what to do to protect ourselves if we did. I think it was. ” He added that now is the time to pass legislation to ensure the ethical use of AI.

Global cooperation is needed now

As AI continues to advance at breakneck speed, even surpassing the expectations of its creators, our ability to steer this technology in directions that are beneficial to humanity becomes increasingly difficult and critical. Governments, businesses and civil society must support collective and collaborative action to overcome local concerns in order to quickly find an ethical and sustainable path forward.

Inclusive and global governance of AI is urgently needed. Getting this right can be important. The future of humanity may depend on how we approach and respond to the challenges of advanced AI.