Artificial General Intelligence

Beyond human intelligence

A world with physical constraints

We start a project to solve one or multiple problems. To make something we need. We faced, face and will face climatic challenges, health diseases, economic crises and political instabilities. We need sustainable growth and increased capacities in research and economical strategies. Multiple limits have been pushed back by energy and resources. We wouldn't have the society we have without iron, oil, uranium, etc.
Last century seemed to be limitless.
This century seems to be limited.
We see limits on growth, limits on resources, on their exploitation and climatic limits. We need to outsmart these limits. We need specific approaches, for each limit, but we also need global approaches, to help push every limit.
We also know that more intelligence helped us do what we wanted to do.

A solution

Base

Artificial General Intelligence (AGI) is a software with all intellectual capacities at least at the level of a human being. Recent advances in AI produce very powerful algorithms, beyond human capabilities, but which are extremely efficient on one specific task. They could be named "Artificial Specialized Intelligences".
If our sustainability is bounded by physical limits, we need to expand our intellectual capacities. As human beings, we are limited by the duration of our studies, ageing, fatigue, limited brain etc. The world is limited, but Humans also have physical limits. Softwares do have physical limits, but these limits are evolving and getting corrected faster than ours. Hardware is getting better, we can optimize computations, multiply computational power by the number of devices etc.

Capacities

We know that being as smart as a human being is possible, because we exist. Though we don't know if a software can be. Making this software would already have some applications, depending on the hardware cost to run it. But what matters the most is the possibilities it could open.

Possibilities

Ok, we make one AGI, and then what? An AGI is at least as smart as a human being. Human beings are great to push limits and solve problems, but having one more human being won't do much.
Now, what's going to happen?
Software is duplicable for free, so we only need more hardware.
If this AGI run at our speed on one unit of hardware, maybe it'll run twice as fast on two units?

  • Maybe it'll run 100 times faster on 100 units?
  • Maybe it'll run 10,000 times faster on 100 units with its based code optimized to run 100 times faster?
  • Maybe it'll run 1,000,000 times faster if above all of that, hardware gets 100 times better?
One day for you to learn something, 2739 years for an AGI. If this is a possibility to solve problems, how fool would we be to not try it?

Limits

We don't know if anything can outsmart the smartest human being, as it has never been done.

  • Maybe the hardware to run an AGI would cost billions of dollars, if possible at all.
  • Maybe the software couldn't use parallel computing.
  • Maybe having a human as smart as a 20 years old adult would require 20 years of training.
  • Maybe there are limits we're not aware of on human intelligence.
  • Maybe we need brain-like hardware.
  • Maybe our society will face extreme challenges sooner than expected.
There are technical limits bound with cognitive sciences limits.

Means

AGI isn't new. It's an old song. Deepmind was founded in 2010. OpenAI in 2015. These are just the most famous ones because they're supported by big structures. But before that, Turing already theorized artificial intelligence in 1950, and before him there's a long road of little steps toward that goal. Did anything change at all? How would we do it now as no one did it before?

We have our skills, our will, and our conviction that AGI algorithms can help solve the challenges of this century.

Development period

As the means seem to have never been that developed, if it's feasible now, it should be done quickly. Some companies and research labs have been working on it for ten years and more, and they've made a lot of progress.
They did achieve things that seemed impossible.
Theoretical or practical advances to progress in virtual worlds need to be complemented by approaches that evolve within the full range of real-world possibilities. If it's feasible now, we think it'll be developed in less than 5 years. Otherwise, we probably lacked big chunks of theory in cognitive sciences, or computational power.

Success rate

Every project may fail.
If we learn that it's impossible to do it with deep learning, it would be a success.
If we advance research, just like others did, it will be a success.
But the purity of research isn't always an end in itself. On the path to Artificial General Intelligence, we are sure we can provide real world usage for standard Deep Learning (DL) algorithms. As our goal is to build AGI from DL, we couldn't do it without making more standard algorithms for every business and people's needs.
It's a 100% success rate from this point of view.
Of course, as AGI has never been done, and already tried, it's a 0% empirical success rate.
We're pretty much in between.

News

  • Card image
    Simulons des pandémies

    Propagation d’un virus dans une population Cet article a pour but de transmettre un retour d’expérience sur la simulation de pandémies. Objectif? Comprendre les limites et l’intérêt de ces a...

    Thu, 22 Oct 2020 11:56:53 GMT

    Read

    Thu, 22 Oct 2020 11:56:53 GMT

    Read
  • Card image
    Neural Network in C++ From Scratch and Backprop-Free Optimizers

    In this article I’ll present a beginner-oriented framework implementing neural networks in C++. The main goal of this code is to understand the root of neural networks for beginners, it also allows ...

    Tue, 13 Oct 2020 23:03:00 GMT

    Read

    Tue, 13 Oct 2020 23:03:00 GMT

    Read
  • Card image
    A more parameter-efficient SOTA bottleneck! (2020/07)

    Linear Bottleneck with Efficient Channel Attention instead of Squeeze Excitation CNN are great blablabla… Let’s get to the point. SOTA for image classification on Imagenet is EfficientNet with 88....

    Sat, 25 Jul 2020 19:40:43 GMT

    Read

    Sat, 25 Jul 2020 19:40:43 GMT

    Read
  • Card image
    Visuels de l’apprentissage des réseaux de neurones

    Modification des représentations internes d’un réseau de neurones en cours d’entrainement 1. Le formalisme de l’apprentissage automatique L’apprentissage automatique est la science regroupan...

    Wed, 03 Jun 2020 19:07:53 GMT

    Read

    Wed, 03 Jun 2020 19:07:53 GMT

    Read
  • Card image
    Construire un serveur de Deep Learning en 2020

    [UPDATE 2020/10/03: Prise en compte des nouveaux GPUs de Nvidia] L’intelligence artificielle, à travers l’apprentissage profond, est une discipline bien établie. Les algorithmes utilisés progr...

    Sat, 16 May 2020 12:43:19 GMT

    Read

    Sat, 16 May 2020 12:43:19 GMT

    Read
  • Card image
    Première Application: AI Magic Mirror !

    Notre première application est disponible ici: https://www.hyugen.com/project/ap_face Il s’agit d’une application de mesure de beauté sur webcam ou fichier. La métrique de beauté est principal...

    Sat, 09 May 2020 23:23:01 GMT

    Read

    Sat, 09 May 2020 23:23:01 GMT

    Read
  • Card image
    Le projet: L’intelligence Artificielle Générale

    Le problème: Les limites Nous démarrons un projet pour résoudre un ou plusieurs problèmes. Pour concevoir une réponse à un besoin. Nous avons fais, faisons et allons faire face à des défis cl...

    Sat, 09 May 2020 16:11:36 GMT

    Read

    Sat, 09 May 2020 16:11:36 GMT

    Read
  • Card image
    Lancement du site web !

    Aujourd’hui nous lançons le site web autour de la (future) société Hyugen qui aura pour but la conception et l’utilisation d’une intelligence artificielle générale. Le site web servira à p...

    Sat, 09 May 2020 15:44:08 GMT

    Read

    Sat, 09 May 2020 15:44:08 GMT

    Read