Life is like a game - DeepMind: The Podcast (S1, Ep3)

Selected as “New and Noteworthy” by Apple Podcasts, the highly-praised, award-nominated first season of "DeepMind: The Podcast" explores the fascinating world of artificial intelligence (AI). Join mathematician and broadcaster Hannah Fry as she meets world-class scientists and thinkers as they explain the foundations of AI, explore some of the challenges the field is wrestling with, and dives into the research that's led to breakthroughs like AlphaGo and AlphaFold. Whether you’re a beginner or an experienced researcher, join our journey into the past, present, and future of AI.
In episode 3, Hannah explores why video games have become a favourite tool for AI researchers to test the abilities of their systems. She hears how systems learned to cooperate in a digital version of the favourite playground game, “Capture the Flag,” and dives into the challenging environment of StarCraft II, a video game that requires players to control the onscreen action with as many as 800 clicks per minute, and how DeepMind’s program AlphaStar learnt to play the game - even beating a top professional player.
Listen in as research scientists explain collaboration, decision-making, and rich environments in the world of games. 🎮
“StarCraft is one of the most complicated games we’ve ever tackled, it’s challenging our understanding, and our algorithms”. - Oriol Vinyals, Research Scientist at DeepMind
#AlphaStar #collaboration #decisions #gaming #neuralnetworks #DMpodcast
_ _
Listen to the full series on UA-cam: dpmd.ai/3geDPmL
Or search for “DeepMind: The Podcast” on your favourite podcast app, including:
Search “DeepMind: The Podcast” and subscribe on your favourite podcast app.
Apple Podcasts: dpmd.ai/2Rzlmcu
Google Podcasts: dpmd.ai/3geDjp5
Spotify: dpmd.ai/3w29cb4
Pocket Casts: pca.st/30m1

[@pauljackson7884]

Great podcast series. Keep them coming please... such an important time in our evolution.

[@kgabhi]

Absolutely fascinating!

[@MathPhilosophyLab]

So great!

[@axiom6919]

RIP John Conway

[@wizardOfRobots]

Where are the shownotes?

[@BagerAkbay]

do you know any texts about a generalized fitness function of intelligence?

[@vegahimsa3057]

You'd have to define "intelligence" in terms of goals. More specifically: a value function of win/loss reward.

[@BagerAkbay]

@@vegahimsa3057 a general fitness function of an entity, a function which might be used for a society, a human being, for a tomato or a bacteria.

[@vegahimsa3057]

You'd still have to define a goal. Intelligence is neither a goal nor even well defined. As for evolution, the "winning fitness strategy" seems to only be whatever maximizes reproduction. I'd argue that intelligence, as it's roughly defined, is not alone optimal and in most cases detrimental. Which species knows that it's destroying its own habitat and accelerating the process? Which of the species has most children?

[@BagerAkbay]

@@vegahimsa3057 Thank you very much for the answer. I know it is not well defined and my aim is trying to find a definition. That's the reason I am asking the question. Specie based winning! condition is not in my interest. I am thinking more about how systems define their goals in their life time and what is the main goal which derives sub-goals. I will try to summarize in a blog post and link it here.

[@vegahimsa3057]

You could arbitrarily define intelligence as: defining goals that maximize fitness (long term survival of individual or species or genome and reproduction) and then most effectively executing the optimal strategy... But humans don't even do anything close to that, so it's a definition of intelligence that no human should accept. Rather humans perceive some version of the world, interact with it, perceive immediate value (emotive valence: pain, pleasure, ambivalence), learn, and act to maximize (immediate) valence. We learn that long term goals, skills, etc create immediate positive valence. All addiction, low intelligence, and mental illness can be defined as a poor evaluation function against a set of "normal" goals.
AlphaGo is intelligent ONLY in terms of winning the game of go. Alpha Zero is intelligent ONLY in terms of whatever it's able to learn and master. Surely an intelligence should be able to master any and many simultaneous tasks. But (to answer your question) we don't yet know how to inform an AI as to WHAT it should learn and master. In fact, we're not even sure that would be ethical.
Suppose you created a robot with an hour of battery charge and switch it on. It should learn (somehow, maybe by its "mother") that it needs to secure an energy source, defend itself, ensure it can continue to obtain an energy source in the future, repair itself, ... And, since you're talking fitness: reproduce (program and build) similar or better versions of itself.
Here intelligence and fitness are defined in terms of survival and reproduction. I'm not sure that's really what humans want out of their AI though. Rather, it's what we fear.

[@elmercodling3989]

Artificial Stupidity is way ahead.

[@Joel-ln9kw]

Finally on my X team

[@vitocarlito5393]

The strong will survive but the best one God takes first.

[@vitocarlito5393]

In life I have one (1) life in game I can do it over and over again. (&)

[@zalxder]

I've been battling AI my entire life #borntogame

[@Propherex]

I watched these StarCraft games and I'm sorry to say that, but it's not clear for me, that AI won "in right way".
TLO played the protoss weakly, which is not surprising, because this is not his main race. Against MaNa, AI won only thanks to the "inhuman" micro (also in TLO games). 600APM on zerg is not equal to 600 of perfect APM on AI protoss.

[@alfonso6263]

Thing is AlphaStar never spams. It only has EPM, something that no one (even pro players) cares, prioritizing their high fake APM most of the time. TLO is a good e.g. The most high fake APM and the less efficient player at the same time, even with zerg.

[@Matlockization]

Images would have been nicer.

[@uydudanbak]

Need beyond ai

[@wriveros]

Synthesized speech

[@zweivessaliuz4402]

.

[@zweivessaliuz4402]

#Hence, we pray No Game, No Life.

[@user-pi8bf6sr4r]

Sooner or later, there will be second Luddite movement.
How can we win this AI?
Is is possible?
And finally AI is better than human in every part of work
What society system do we need??
Capitalism Again?

[@Wagmiman]

Stop spamming