Home » Technical Topics » Knowledge Engineering

Watching the Shift Towards More Symbolic AGI

  • ajitjaokar 
Watching the Shift Towards More Symbolic AGI

For a long time, Gary Marcus was a lonely voice in favour of symbolic(more specifically hybrid) AI models

Recent events seem to have emboldened his view

The debate is significant because it has implications for the future of AI


  • From the dawn of AI, the debate between symbolic vs neural network approaches of AI has continued
  • From 2010 onwards, the emphasis has shifted to the neural network approaches led by Hinton, LeCun and Bengio
  • We also have a middle ground: i.e.  “hybrid models” that try to combine the best of both worlds, by integrating the neural networks with the powerful abstraction capacities of symbol manipulation.
  • In response to an article from Yann LeCun, Gary Marcus argues that in fact, the neural network community is now (grudgingly) accepting the symbolic viewpoint
  • This blog summarises the Gary Marcus perspective on why the neural network community seems to be indirectly accepting the symbolic mindset – while yet not explicitly acknowledging symbolic approaches
  • The debate is significant because it has implications for the future of AI
  • According to Gary Marcus, in a recent essay while Yann LeCun, seems to reject hybrid models, by the end of the essay, seems to acknowledge that hybrid systems exist and are possible.   
  • The crux of the argument seems to be that LeCun and Browning believe that a model isn’t hybrid if it learns to manipulate symbols. 
  • But for Gary Marcus, it is an acknowledgement that the neural network community is shifting its stance towards the significance of hybrid models. 


  • Early AI pioneers like Marvin Minsky and John McCarthy took the symbolic approach but others like Frank Rosenblatt favoured the neural network approach
  • There were also proposals that the two possibilities aren’t mutually exclusive.
  • More recently, the separation has been more acute  with LeCun, Hinton and Yoshua Bengio proposing that most problems can be solved with the neural network approach alone ex Hinton quote that  “Deep learning is going to be able to do everything.”

However, the sands of time are shifting it seems as per Gary Marcus 

“The fact that LeCun is even considering a hypothesis that embraces symbol manipulation, learned or otherwise, represents a monumental concession, if not a complete about-face. Historians of artificial intelligence should in fact see the Noema essay as a major turning point, in which one of the three pioneers of deep learning first directly acknowledges the inevitability of hybrid AI. “

And there are other examples such as Andrew Ng , Jürgen Schmidhuber’s AI company NNAISENSE and Yoshual Bengio discussion on “System 2” cognition  

To conclude, my analysis is that these discussions are good for the industry and propel us forward as a whole in the spirit of scientific discussion

Reference Gary Marcus – Deep learning alone isn’t getting us to human-like AI

Image source Pixabay