Adapting the Process of Adaptation: Darwinian, Lamarckian, and Agent-Directed Evolution in AI
Organizations: MIT Sloan School of Management; MIT Laboratory for Financial Engineering; MIT Computer Science and Artificial Intelligence Laboratory; Santa Fe Institute; 100 Main Street, E62–618, Cambridge, MA 02142 · MIT Sloan School of Management; MIT Laboratory for Financial Engineering. 1 Main Street, E90, Cambridge, MA 02142.
Abstract
Evolution has shaped intelligence for billions of years. Artificial intelligence (AI) now raises a striking possibility: intelligence may begin to direct its own evolution. As AI agents become capable of modifying and evaluating their own behavior, how quickly could they improve, and how can that improvement remain safe? We develop a theoretical framework comparing Darwinian, Lamarckian, and agent-directed evolution following environmental change. Darwinian evolution relies on natural selection, Lamarckian evolution adds inherited learning, and agent-directed evolution allows agents to generate and evaluate behavioral variants and choose how extensively to search. When learning is sufficiently effective, these mechanisms exhibit progressively faster adaptation. Generating more variants accelerates adaptation, while optimal search effort depends on learning ability, problem difficulty, and experimentation costs. However, an agent's ability to direct its improvement does not ensure that improvements are evaluated against the intended objective. When the evaluation target is misspecified, additional search accelerates convergence toward the wrong behavior, even as measured performance improves. In recursive self-improvement, such errors may persist across generations, posing challenges for AI safety. Independent evaluation and oversight must keep pace with autonomous AI development. The central challenge is not simply to accelerate the evolution of intelligence, but to preserve humanity's ability to guide its direction, limit its speed, and intervene when necessary.
Figures & tables
| Agent-directed | ||||||
|---|---|---|---|---|---|---|
| Remaining regret | Darwinian | Lamarckian | ||||
| 10% | 125 | 11 | 3.4 | 2.7 | 2.2 | 2.0 |
| 5% | 130 | 15 | 4.3 | 3.4 | 2.9 | 2.7 |
| 1% | 145 | 22 | 6.2 | 4.9 | 4.1 | 3.9 |
| 0.1% | 160 | 33 | 9.0 | 7.1 | 5.9 | 5.2 |
Appendix figures & tables5 assets
Supplementary material from the paper’s appendix.
Appendix
| Symbol | Meaning | Introduced |
|---|---|---|
| Behavior of a lineage in generation : the probability with which each of its individuals chooses action | Section 3 | |
| State of the environment when generation is formed from generation , equal to with probability and with probability after the shift | Section 3 | |
| , | Offspring of an individual choosing action and of one choosing action : for the action that succeeds and for the other | Section 3 |
| Offspring produced by the successful action | Section 3 | |
| Probability that action succeeds; also the optimal behavior | Section 3 | |
| Size of the population in generation | Sections 3.1 and 3.2 |
| This paper | Evolutionary biology | AI systems |
|---|---|---|
| Inherited behavior | Heritable characteristics | Inherited model or agent configurations |
| Generation | Reproductive cycle | Model or agent improvement cycle |
| Variant | Offspring | Modified models or agents |
| Growth rate; growth regret | Reproductive fitness; fitness shortfall | Performance; gap from optimal performance |
| Darwinian evolution | Natural selection | Selection among existing model or agent variants |
| Lamarckian evolution | Inheritance of acquired characteristics | Learned improvements inherited by subsequent models or agents |
Self-Evolving Agents
Self-evolving agents improve their skills, memory, tools, policies, or environments by learning from interaction and evaluation feedback.
See all 447 papers