AI research endured years of failure and frustration before new techniques in deep learning unleashed the swift, astonishing progress of the last decade.

Michael’s recent book A Brief History of Artificial Intelligence explores what we can learn from this history, and examines where we are now and where the field is going.

We discuss:

• Why the Cyc project’s aim to encode ‘all human knowledge’ (!!) into a functioning AI got stuck, despite years of intense effort. 
• What OpenAI’s GPT-3 language-generating AI system really knows about making an omelette.  
• Why the next generation of AI systems may have to combine symbolic and non-symbolic approaches. 

You can find NOUS on Twitter @NSthepodcast 

The idea we have ‘innate knowledge’ seems quite wrong to most of us. But we do! And the intuitions leading us astray here also blind us to other aspects of human nature. 

We are all ‘blind storytellers’. Professor Iris Berent reveals what misleads us, and what we are missing.

18:55 Newborns have basic knowledge of the nature of objects. Eye-tracking experiments reveal that they have a grasp of the 3 c’s - cohesion, contact and continuity. 

22:35 How do you get expectations about the nature of the world coded into genes? Do genes somehow give rise to computational ‘rules’ in the brain? Is my inability to grasp this illustrating Iris’ argument!? A deep mystery remains. 

26:51 Birdsong is innate. So why not aspects of language and human object cognition?  

28:20 “People know how to talk in more or less the sense that spiders know how to spin webs“ says Steven Pinker. 

37:44 We learn a particular language from those around us - but some argue that the deep structural rules underlying all languages are innate. How does that work? Are there ‘rules’ of language somehow inscribed in neural structures? 

47:39 Our intuitive biases to *dualism* and *essentialism* lead us to get lots of things wrong about human nature. 

55:05 Why we go ‘insane about the brain’, and get weirdly impressed by neuroscience-y explanations, even when they are bad. 

1:00:44 Why is our thinking about mental disorders so biased and confused? 

***

Check out Iris Berent's book The Blind Storyteller here, or find her on Twitter @berent_iris

To get in touch with Ilan or join the conversation, you can find NOUS on Twitter @NSthepodcast or on email at nousthepodcast@gmail.com

Vision is the best understood sensory domain. But smell is turning out to be wonderfully strange and even more complex than sight.

Dr Ann-Sophie Barwich joins me to explore ideas from her recent book Smellosophy. How is vomit related to parmesan cheese? Why do things smell so different depending on context? And what does smell teach us about the very nature of perception? 

We explore:

• Why the ‘promiscuity’ of smell doesn’t make it merely subjective. Smells can have a multitude of qualities or notes depending on the context and depending on the individual. But this variability has a functional basis. 
• The weird neural representation of smell. The patterns of neural activation underpinning smell don’t follow the mapping principles followed by other sensory modalities. 
• Why philosophers shouldn’t ignore the neural ‘plumbing’ of sensory systems. Evolved brain mechanisms underly the nature and function of the perceptual experience  - so they have to inform a philosophical account of perception.  

Check out Dr Barwich’s book Smellosophy here, https://www.hup.harvard.edu/catalog.php?isbn=9780674983694, 

her article in Aeon magazine here https://aeon.co/essays/why-might-it-be-easier-to-fool-your-eyes-than-your-nose.

and another great piece in Nautilus http://nautil.us/issue/91/the-amazing-brain/our-mind_boggling-sense-of-smell 

***

To get in touch with Ilan or join the conversation, you can find NOUS on Twitter @NSthepodcast or on email at nousthepodcast@gmail.com