Welcome

Evo­lu­tion may be more intel­li­gent than we thought, accord­ing Pro­fes­sor Richard Wat­son from the Uni­ver­sity of Southamp­ton in the United Kingdom.

Pro­fes­sor Wat­son says new research shows that evo­lu­tion is able to learn from pre­vi­ous expe­ri­ence, which could pro­vide a bet­ter expla­na­tion of how evo­lu­tion by nat­ural selec­tion pro­duces such appar­ently intel­li­gent designs.

By uni­fy­ing the the­ory of evo­lu­tion (which shows how ran­dom vari­a­tion and selec­tion is suf­fi­cient to pro­vide incre­men­tal adap­ta­tion) with learn­ing the­o­ries (which show how incre­men­tal adap­ta­tion is suf­fi­cient for a sys­tem to exhibit intel­li­gent behav­iour), this research shows that it is pos­si­ble for evo­lu­tion to exhibit some of the same intel­li­gent behav­iours as learn­ing sys­tems (includ­ing neural networks).

In an opin­ion paper, pub­lished online on 16 Decem­ber in Trends in Ecol­ogy and Evo­lu­tion, Pro­fes­sors Wat­son and Eörs Sza­th­máry, from the Par­menides Foun­da­tion in Munich, explain how for­mal analo­gies can be used to trans­fer spe­cific mod­els and results between the two the­o­ries to solve sev­eral impor­tant evo­lu­tion­ary puzzles.

Pro­fes­sor Wat­son says: “Darwin’s the­ory of evo­lu­tion describes the dri­ving process, but learn­ing the­ory is not just a dif­fer­ent way of describ­ing what Dar­win already told us. It expands what we think evo­lu­tion is capa­ble of. It shows that nat­ural selec­tion is suf­fi­cient to pro­duce sig­nif­i­cant fea­tures of intel­li­gent problem-​solving.”

For exam­ple, a key fea­ture of intel­li­gence is an abil­ity to antic­i­pate behav­iours that that will lead to future ben­e­fits. Con­ven­tion­ally, evo­lu­tion, being depen­dent on ran­dom vari­a­tion, has been con­sid­ered ‘blind’ or at least ‘myopic’ — unable to exhibit such antic­i­pa­tion. But show­ing that evolv­ing sys­tems can learn from past expe­ri­ence means that evo­lu­tion has the poten­tial to antic­i­pate what is needed to adapt to future envi­ron­ments in the same way that learn­ing sys­tems do.

“When we look at the amaz­ing, appar­ently intel­li­gent designs that evo­lu­tion pro­duces, it takes some imag­i­na­tion to under­stand how ran­dom vari­a­tion and selec­tion pro­duced them. Sure, given suit­able vari­a­tion and suit­able selec­tion (and we also need suit­able inher­i­tance) then we’re fine. But can nat­ural selec­tion explain the suit­abil­ity of its own processes? That self-​referential notion is trou­bling to con­ven­tional evo­lu­tion­ary the­ory — but easy in learn­ing theory.

“Learn­ing the­ory enables us to for­malise how evo­lu­tion changes its own processes over evo­lu­tion­ary time. For exam­ple, by evolv­ing the organ­i­sa­tion of devel­op­ment that con­trols vari­a­tion, the organ­i­sa­tion of eco­log­i­cal inter­ac­tions that con­trol selec­tion or the struc­ture of repro­duc­tive rela­tion­ships that con­trol inher­i­tance — nat­ural selec­tion can change its own abil­ity to evolve.”

(Source: Uni­ver­sity of Southamp­ton media release, 18.12.2015)