Ratcheting evolution

Ratcheting Evolution,

According to the Biology

(Algorithms for Evolution)Joshua Knowles - [email protected], with heavy

debts to Richard A. Watson (U. Southampton)

DarwinCharles Darwin is widely regarded as the forefather of Evolution by Natural Selection

His theory is the cornerstone of much biology, and is at the centre of Evolutionary Biology – the study of how genes, organs, organisms, and on all the way up to ecosystems evolve over time.

Let’s applaud Darwin!

The Evolutionary Synthesis

Modern evolutionary theory admits that while Darwin was right on many, many things, he did not know genetics (or genetics had not been discovered)

So, the “Modern Synthesis” of how evolution works adds Mendelian genetics to Darwin, and adds some other things too.

Slide by Tony Malone

Extending the Evolutionary Synthesis

More recently, new and persistent biological evidence has been emerging, which suggests that there are many mechanisms that genes, proteins, cells, organs, organisms and whole ecosystems use to evolve.The mechanisms are• Coevolution • Niche construction / speciation• Phenotypic plasticity• Gene reordering• “Junk” DNA• Major transitions, and symbiogenesis – Margulis• GAIA theory (the whole world is an organism)These are not really part of the Darwinian model ;-(


YOU: OK, Josh, but what the heck has that got to do with computer science?



ME: Well, natural selection is an algorithm. That is to say, we can run a simulation of natural selection on our laptops and evolve all sorts of cool stuff ...

N.B.: When we run a simulation of natural selection, this isnatural selection, according to most philosophers. It is justthe same as saying that adding 2+2 on paper is exactly the same as taking two objects and putting them next to two others.




ME: And what’s more, those algorithms could be improved greatly if we added in more ways of evolving. We need to get closer to the new biology, but without the detail.

This is easily said!

Evolutionary AlgorithmsThe products of evolutionary algorithms range over

BridgesMilitary strategiesArtificially intelligent game-players, and bots on the Internet(Physical) Swarm robots, and how they move and cooperateTelephone and computer network designsRoasting recipes for making chocolate!Drug designs

...and many more things besides.


A drug evolved from DNA![see my papers]

An antenna evolved for a NASA space probe


Automated mass spectrometry,as seen in Batman [see my papers on this]

Roasted cocoa beans with wonderful aroma[see my papers]


Work from the 1970s done by Hans-Paul Schwefel and colleagues in Berlin

Evolutionary algorithms: the algorithm

Q. How does it work?

Answer: 1. Decide what you want to evolve2. Choose an appropriate measure of quality3. Start with some random attempts4. Select the best few, according to your quality measure5. Create new copies of these few, but with some small

changes. Let the others die ;-(6. Keep going back to step 4. until you are happy!

N.B. it is crucial that step 5 allows inheritance of most of the characteristics, unchanged.




ME: And what’s more, those algorithms could be improved greatly if we added in more ways of evolving. We need to get closer to the new biology, but without the detail. This is easily said!


Coevolution

Niche construction (speciation)

Lifetime learning (heritable)

Symbiogenesis! – a theory by Lynn Margulis

GAIA!!! – a theory by James Lovelock

These are just some of the ways we could add stuff !


Coevolution

Niche construction (speciation)

Lifetime learning (heritable)

Symbiogenesis! – a theory by Lynn Margulis

GAIA!!! – a theory by James Lovelock

These are just some of the ways we could add stuff !

Heritable lifetime-learning

We adapt, during life – that is clear

How much of the adaptation is inherited? – This has long been controversial

HOW could things learned in your brain (or heart) be passed on to the next generation via your genes?!?

Conventional (Darwin) view: they can’t

Maverick (Shapiro) view: they can!! ... So there!We want to create algorithms to answer the HOW, and give a bit of a better justification than “So there!” ;-)

We’ve made a good start

Richard A. Watson’s work made the front cover of New Scientist, and was in the Independent newspaper, both in 2016.

We’ve made a good start

Joshua Knowles’s work on evolutionary algorithms has been cited in15,000 scientific papers in just 15 years. And his recent NIPS slides wereviewed online >400 times in less than two weeks

Holmes and Watson...solved a few mysteries.

So, what about Knowles and Watson?Or, Watson and Knowles?

Watch THIS SPACE !

Holmes and Watson...solved a few mysteries.

So, what about Knowles and Watson?Or, Watson and Knowles?

Watch THIS SPACE ! And are YOU going to help us?

Selected references (very selected)

Joshua Knowles (2006) “ParEGO:...”, IEEE Transactions on Evolutionary Computation (cited >400 times; IEEE award-winning)

Richard A. Watson (2016) “Why evolution may be smarter than we thought”, The Independent

Our publications full listings are online at these two places:goo.gl/FMCXIA - Joshua Knowlesgoo.gl/hIbyH9 - Richard A. Watson

Ratcheting evolution

Education