Main uses of evolutionary algorithms
- Maintaining a population of different solutions
- Optimizing a single solution.
MORE IDEAS FROM THE ARTICLE
Pursuing specific goals can be a hindrance to reaching those objectives.
Kenneth Stanley, a computer scientist, hoped to show that by following ideas in interesting directions, algorithms can produce a diversity of results and solve problems. Thus, ignoring an objective can get you to the solution faster than pursuing it. He showed this through an approach named novelty search.
Traditionally, evolutionary algorithms are used to solve specific problems. For instance, the ability to control a two-legged robot. Solutions that perform the best on some metrics are selected to produce offspring.
In spite of successes, these algorithms are more computationally intensive than approaches such as "deep learning."
Neuroevolution is a form of artificial intelligence. It is a meta-algorithm, an algorithm for designing algorithms. It adopts the principles of biological evolution in order to design smarter algorithms. Eventually, the algorithms get pretty good at their job.
Biological evolution is the only system to produce human intelligence.
If we want algorithms that can navigate the physical and social world as we do, we need to imitate nature's tactics. Instead of hard-coding for specific metrics, we must let a population of solutions blossom that may discover an indirect path or a set of stepping stones to allow them to evolve better than if they'd received those skills directly.
It goes beyond traditional evolutionary approaches. It explains innovation. Instead of optimizing for a specific goal, it embraces the creative exploration of a diverse population of solutions.
The steppingstone’s potential can be seen by analogy with biological evolution: feathers likely evolved for insulation and only later became handy for flight.
- It requires a considerable amount of computation.
- Most people want to solve a particular problem. Neuroevolution might get you to a creative solution but could lead you astray before it puts you on the right path.
Pros
- Other AI, like reinforcement learning, can lead to short-terms gains, but will eventually get stuck in a rut. Random exploration in the steppingstone principle can beat the best algorithms.
- Open-ended exploration might lead to human-level intelligence via paths we would never have anticipated.
RELATED IDEAS
As machines become increasingly capable, along with computer memory, power and space being abundantly available, our brains are in a transition phase.
Earlier we had to remember a lot, do calculations on paper, and jog our memories to recall something. More and more of such information isn't processed by our brains anymore and is taken care of by machines.
How much can we afford to forget, if we train machines to remember? - Gene Tracy | Aeon Ideas
aeon.co
Although AI researchers can train systems to win at Space Invaders, it couldn’t play games like Montezuma Revenge where rewards could only be collected after completing a series of actions (for example, climb down ladder, get down rope, get down another ladder, jump over skull and climb up a third ladder).
For these types of games, the algorithms can’t learn because they require an understanding of the concept of ladders, ropes and keys. Something us humans have built in to our cognitive model of the world & that can’t be learnt by the reinforcement learning approach of DeepMind.
Artificial intelligence (AI) has been around since the 1950s. The original pioneers dreamed of a computer that could perform tasks like humans, such as playing chess or translating languages. But the plans didn't come to fruition, and AI soon fell out of favour.
AI technology continued to improve exponentially over the next few decades. Many organisations now embrace AI as a core element of their business.
❤️ Brainstash Inc.