Google DeepMind’s AI Agent Dreams Up Algorithms Beyond Human Expertise
1 min read
Summary
DeepMind’s AlphaEvolve project involves an AI model that is able to surpass humans in designing algorithms for various real-world problems.
The model uses an evolutionary method to generate novel solutions and involves testing each iteration until a better algorithm is born; it can even design more efficient matrices for calculations.
The coding agent can take on tasks in AI, centre scheduling, chip design and many others, according to the head of AI for science at DeepMind, Pushmeet Kohli.
Whilst it is difficult to know if a piece of code is entirely novel, the model has been designed in such a way that the programmers can be sure that the output is both correct and new.
The potential for the project is mammoth, as it demonstrates an ability to generate ideas and solutions beyond those originally programmed.
This could have huge implications in the future of science and innovation.
