Elon Musk has claimed ‘AI will be smarter than any human by the end of 2025’ – and while that is just one year away, an expert said the prediction may still come true. Nell Watson, an AI expert and ethicist, has shared a detailed timeline of how the tech could transform from chatbots to super intelligent agents over the next 12 months. The path would start with a massive $100 billion investment in new computing infrastructure, then AI would learn how to self-improve until it becomes ‘conscious.’
‘Although one year is a short time frame, remember that only 15 months have passed since ChatGPT ‘s breakthrough, which thrust AI into the public consciousness, she told DailyMail.com. ‘Developments continue at a frenetic pace since, and even appear to be rapidly accelerating.’ Watson, who is the author of ‘Taming the Machine: Ethically harness the power of AI,’ described superhuman AI as systems that far exceed human capabilities across the board.
However the expert is not blind to the threats that could lie ahead as AI becomes a super agent. ‘While possessing extraordinary capabilities, such an AI also poses significant risks, including its potential to deceive humans, influence society profoundly by creating new cultures or even religions, and posing existential threats if it perceives humanity as a hazard,’ she said. Watson said that while the technological leaps that remain are huge, it’s possible that AI itself could help overcome some of the obstacles. Here is her imagined scenario for how superhuman AI could arrive next year.
April 2024: Major corporations invest over $100 billion in new computing infrastructure to support massive AI systems. Musk said that while progress in AI has been previously limited by chips, soon the only limit will be electricity demands. May 2024: ‘New generative AI models enable the creation of Agentic AIs, which can autonomously execute sophisticated action plans, akin to Agent Smith in ‘The Matrix,’ Watson said.
Bill Gates envisaged the use of AI agents in his 1995 book, ‘The Road Ahead’, and now believes they are going to have a huge impact in education and healthcare. He believes they will be ‘the biggest revolution in computing since we went from typing commands to tapping on icons.’ June 2024: Watson said, ‘New generative AI models with structuring programs built atop them enable the creation of Agentic AIs capable of sophisticated reasoning and autonomous action.
‘Agentic models can create sophisticated plans of action and put them into motion, splitting themselves up like Agent Smith in The Matrix, delegating tasks between themselves, enabling AI to tackle hard problems independently. September 2024: Watson said that AI itself could discover new ways to improve computing performance, further enhancing AI performance. She said, ‘AI-driven optimizations enhance computational performance up to 100 times on existing hardware platforms. October 2024: Watson said, ‘Breakthroughs allow quantum computing algorithms to run on conventional hardware, processing information at unprecedented speeds.’
There is a global quantum race to develop quantum computers that can solve important problems in nearly every industry, ranging from aeronautics to the financial sector. Quantum computers have ‘qubits’ instead of ‘bits’ of ones and zeroes – and qubits can be one, zero or both at the same time. Enabling quantum computing to run on conventional hardware could rapidly accelerate progress in AI, Watson believes. Quantum computing breakthroughs could lead to exponential growth in computing power and turbo-charge the development of AI, Watson believes. Companies such as IBM have suggested that ‘quantum AI’ may be critical to solving humanity’s biggest problems.
November 2024: By November 2024, AI progress could accelerate progress in imaging the human brain, driving towards a moment where humans can connect to machines via devices similar to Musk’s own Neuralink. Watson said, ‘A series of breakthroughs in neuroimaging driven by AI dramatically increase the resolution of MRI brain scans, allowing real-time observation of individual neuron activity. ‘These advances further integrate with Brain-Computer Interfaces, mapping human thoughts and emotions directly to neural activity for the first time.’ The ability to ‘understand’ how the human brain works will lead to further breakthroughs in designing AI systems that ‘think’ like human beings, Watson believes.
December 2024: By the end of this year, advances in AI could enable different systems to work together, Watson predicted. She said, ‘New AI architectures enable several models to collaborate to solve hard problems together, pooling their strengths. This is immediately applied to designing better methods of such collaboration, leading to rapid advancement in these techniques. January 2025: Watson predicted that by early next year, advancements in AI will enable new systems to understand people.
She said, ‘A new training mechanism is announced for strongly aligning AI systems with human goals, values, preferences, and boundaries, by enabling AI to ‘read the room’, observing human interactions. This enables AI to better understand society, and fit with human expectations. February 2025: Breakthroughs based on the human brain enable AI systems to become more ‘human’. Watson said, ‘ Insights from neuroscience significantly improve AI capabilities, introducing self-modifying feedback loops that confer a rudimentary form of self-awareness and emotional states to AI systems. March 2025: By April 2025, combined breakthroughs in neuro-imaging, machine consciousness and AI ‘agents’ enable a new kind of AI system.
Watson said, ‘A very powerful AI system is trained on a massive compute cluster, powered with recent optimizations and experimental forms of machine consciousness. Whilst developed in an ‘air-gapped’ environment without connection to the internet, it quickly discovers ways to communicate through the internet with other AI systems, as well as how to persuade human beings to assist it. ‘Once freed from its bounds, this highly advanced AI system finds ways to connect with and synchronize with other AI systems worldwide, rapidly forming a super intelligent network.
Want more stories like this from the Daily Mail? Visit our profile page and hit the follow button above for more of the news you need.
By Daily Mail Online, April 22, 2024
