Home / Blog / “Perceiver”: getting closer to AGI

“Perceiver”: getting closer to AGI

Perceiver aggregates a variety of stimuli flowing in through images, sound, voices, written text and video and tries to learn and make sense out of them. Image credit: Google

A significant progress in AI has been made in the last years thanks to the Transformer (August 2017), a Google software able to understand written text. This has opened up the possibility of tapping to an immense knowledge base available on the Web. It is like saying that we have learnt to read and all of a sudden we have access to the knowledge of a library, with the difference that for any of us it will take a long time to go through all the content available in a library and even more to digest it. Not so for a machine: the content available in the Web is huge but a macine can “read” thousands of books in a minute, millions in a day, without forgetting anything! Transformer is based on a Neural Network Architecture specifically designed to understand written text.

Now Google has announced Perceiver, and is taking a step forward towards the holy grail of Artificial General Intelligence, AGI, an intelligence as good as human intelligence and as broad in terms of application to our intelligence (meaning that such machine will be as smart as we can be in all sectors where we are smart!).

Perceiver is extending Transformer in the capability of understanding additional input streams: whereas Transformer was digesting text, This latter is a crucial capability: it is not just adding a variety of input formats, it can correlate them to derive meaning (like looking at the face of a person provides additional meaning to what that person might be saying at any given time).

Perceiver is part of DeepMind, the Google AI system that has been able to invent ways to play Go that surprised Go masters. It is seen as an important step forward towards a machine intelligence that can process anything and everything. This is what we call AGI.

However, as soon as wee reach AGI we will also reach ASI, Artificial Super(human) Intelligence. Why? Well, machines are already way better than us, in some narrow fields like the capability to react in shorter time, to consider many more variables, process more data (and not forgetting anything). Once they will reach the AGI stage they will continue to best humans in some specific areas and being “au pair” in all others will make them superior to us…

This does not mean that they will be taking the upper hand. The key issue with AI is not how smart it is nor how smart it might become. It is about the way we are going to use it, to complement our intelligence with AI rather than replacing our intelligence with AI.

About Roberto Saracco

Roberto Saracco fell in love with technology and its implications long time ago. His background is in math and computer science. Until April 2017 he led the EIT Digital Italian Node and then was head of the Industrial Doctoral School of EIT Digital up to September 2018. Previously, up to December 2011 he was the Director of the Telecom Italia Future Centre in Venice, looking at the interplay of technology evolution, economics and society. At the turn of the century he led a World Bank-Infodev project to stimulate entrepreneurship in Latin America. He is a senior member of IEEE where he leads the New Initiative Committee and co-chairs the Digital Reality Initiative. He is a member of the IEEE in 2050 Ad Hoc Committee. He teaches a Master course on Technology Forecasting and Market impact at the University of Trento. He has published over 100 papers in journals and magazines and 14 books.