Artificial General Intelligence: Concept, State of the Art, and Future Prospects

Ben Goertzel, Founder of OpenCog, presents a high level summary of the Artificial General Intelligence (AGI) field in his 2014 review article [1]. Here I summarize the paper and then share my conclusions.

Summary

The paper can be broken into 5 primary sections.

1. First he presents the core concepts behind AGI.
2. He then attempts to unravel the complexities of understanding and defining general intelligence.
3. Next a careful consideration of projects in the field yield a succinct categorization of modern AGI methodologies. This is the meat of the paper and the pros/cons analysis for each categorization is particularly elucidating.
4. Then many robust graphs and systems modelling structures which underlay human-like general intelligence are presented.
5. Lastly a consideration of metrics and analysis methods is performed.

In less details, the AGI field encompasses all methodolgies, formalisms, and attempts at creating or understanding thinking machines with a general intelligence comparable to or greater than that of human beings. As can be seen from the previous sentence this is a difficult concept to delineate. In light of this, Goertzel presents many qualitative AGI features that roughly describe the purpose and direction of the field. These features are believed to be accepted by most AGI researchers. After some hand waving he presents what he calls, the Core AGI Hypothesis.

Goertzel ensures his readers that this hypothesis is widely accepted by “nearly all researchers in the AGI community”. This contrasts what has come to be known as narrow AI, a term coined by Ray Kurzweil. Narrow AI is synthetic intelligence software designed to solve specific, narrowly constrained problems [2]. A key feature of AGI which needs to be elaborated is the notion of general intelligence. Many approaches to defining and explaining what it means to be “generally intelligent” are proposed. After considering Psychological and Mathematical characterizations, adaptation and embodiment approaches, and cognitive-architectures, Goertzel admits that no widely accepted definition exists. This, we will see, is a recurring theme amid the AGI community.

After the high level introduction to the scope of AGI a succinct categorization of the mainstream AGI approaches is presented. Goertzel partitions the field into 4 categories.

• Symbolic
  
  The roots of the symbolic approach to AGI reach back to the traditional AI field. The guiding principle for all symbolic systems is the belief that the mind exists mainly to manipulate symbols that represent aspects of the world or themselves. This belief is called the physical symbol system hypothesis.

  FOR:
  Symbolic thought is what most strongly distinguishes humans from other animals; it’s the crux of human general intelligence. Symbolic thought is precisely what lets us generalize most broadly. It’s possible to realize the symbolic core of human general intelligence independently of the specific neural processes that realize this core in the brain, and independently of the sensory and motor systems that serve as (very sophisticated) input and output conduits for human symbol-processing.

  AGAINST:
  While these symbolic AI architectures contain many valuable ideas and have yielded some interesting results, they seem to be incapable of giving rise to the emergent structures and dynamics required to yield humanlike general intelligence using feasible computational resources. Symbol manipulation emerged evolutionarily from simpler processes of perception and motivated action; and symbol manipulation in the human brain emerges from these same sorts of processes. Divorcing symbol manipulation from the underlying substrate of perception and motivated action doesn’t make sense, and will never yield generally intelligent agents, at best only useful problem-solving tools.

• Emergentist
  
  The Emergentist approach to AGI takes the view that higher level, more abstract symbolic processing, arises (or emerges) naturally from lower level “subsymbolic” dynamics. As an example, consider the classic multilayer neural network which is in most ubiquitous practice today. The view here is that a more thorough understanding of the fundamental components of the brain and their interplay may lead to a higher level understanding of general intelligence as a whole.

  FOR:
  The brain consists of a large set of simple elements, complexly self-organizing into dynamical structures in response to the body’s experience. So, the natural way to approach AGI is to follow a similar approach: a large set of simple elements capable of appropriately adaptive self-organization. When a cognitive faculty is achieved via emergence from subsymbolic dynamics, then it automatically has some flexibility and adaptiveness to it (quite different from the “brittleness” seen in many symbolic AI systems). The human brain is actually very similar to the brains of other mammals, which are mostly involved in processing high-dimensional sensory data and coordinating complex actions; this sort of processing, which constitutes the foundation of general intelligence, is most naturally achieved via subsymbolic means.

  AGAINST:
  The brain happens to achieve its general intelligence via self-organizing networks of neurons, but to focus on this underlying level is misdirected. What matters is the cognitive “software” of the mind, not the lower-level hardware or wetware that’s used to realize it. The brain has a complex architecture that evolution has honed specifically to support advanced symbolic reasoning and other aspects of human general intelligence; what matters for creating human-level (or greater) intelligence is having the right information processing architecture, not the underlying mechanics via which the architecture is implemented.

• Computational Neuroscience
  
  As it sounds, computational neuroscience is an approach to exploring the principles of neuroscience using computational models and simulations. This approach to AGI falls under the emergentist category. If a robust model of the human brain can be developed it stands to reason the we may be able to glean insight into what components of the model give rise to higher level general intelligence.

  FOR:
  The brain is the only example we have of a system with a high level of general intelligence. So, emulating the brain is obviously the most straightforward path to achieving AGI. Neuroscience is advancing rapidly, and so is computer hardware; so, putting the two together, there’s a fairly direct path toward AGI by implementing cutting-edge neuroscience models on massively powerful hardware. Once we understand how brain-based AGIs work, we will likely then gain the knowledge to build even better systems.

  AGAINST:
  Neuroscience is advancing rapidly but is still at a primitive stage; our knowledge about the brain is extremely incomplete, and we lack understanding of basic issues like how the brain learns or represents abstract knowledge. The brain’s cognitive mechanisms are well tuned to run efficiently on neural wetware, but current computer hardware has very different properties; given a certain fixed amount of digital computing hardware, one can create vastly more intelligent systems via crafting AGI algorithms appropriate to the hardware than via trying to force algorithms optimized for neural wetware onto a very different substrate.

• Developmental Robotics
  
  Infants are the ultimate scientists. They use all of their senses to interact with their environment and over time create a model of their perceived reality. It is argued that general intelligence arises from “the brain’s” constant interaction with its surroundings and environment. Developmental robotics attempts to recreate this process.

  FOR:
  Young human children learn, mostly, by unsupervised exploration of their environment – using body and mind together to adapt to the world, with progressively increasing sophistication. This is the only way that we know of, for a mind to move from ignorance and incapability to knowledge and capability.

  AGAINST:
  Robots, at this stage in the development of technology, are extremely crude compared to the human body, and thus don’t provide an adequate infrastructure for mind/body learning of the sort a young human child does. Due to the early stage of robotics technology, robotics projects inevitably become preoccupied with robotics particulars, and never seem to get to the stage of addressing complex cognitive issues. Furthermore, it’s unclear whether detailed sensorimotor grounding is actually necessary in order to create an AGI doing human level reasoning and learning.

• Hybrid
  
  In recent years AGI researchers have begun integrating both symbolic and emergentist approaches. The motivation is that, if designed correctly, each system’s strengths can ameliorate the other’s weaknesses. The concept of “cognitive synergy” captures this principle. It argues that higher level AGI emerges as a result of harmonious interactions from multiple components.

  FOR:
  The brain is a complex system with multiple different parts, architected according to different principles but all working closely together; so in that sense, the brain is a hybrid system. Different aspects of intelligence work best with different representational and learning mechanisms. If one designs the different parts of a hybrid system properly, one can get the different parts to work together synergetically, each contributing its strengths to help over come the others’ weaknesses. Biological systems tend to be messy, complex and integrative; searching for a single “algorithm of general intelligence” is an inappropriate attempt to project the aesthetics of physics or theoretical computer science into a qualitative different domain.

  AGAINST:
  Gluing together a bunch of inadequate systems isn’t going to make an adequate system. The brain uses a unified infrastructure (a neural network) for good reason; when you try to tie together qualitatively different components, you get a brittle system that can’t adapt that well, because the different components can’t work together with full flexibility. Hybrid systems are inelegant, and violate the “Occam’s Razor” heuristic.

• Universalist
  
  The univeralist approach leverages a principle employed by many creative designers and inventors. Instead of coming up with an idea that satisfies all of a problems inherent limitations, one “dreams big” and develops elaborate, even unrealistic ideas, and later simplifies them to fit within the confines of the proposed problem. In regard to AGI, the so called universalist approach, aims at developing ideal, perfect, or unrealistic models of general intelligence. These models and algorithms may require incredible power, even infinite power to be employed. In summary universalists might argue that one should not limit their creativity by any imposed constraints.

  FOR:
  The case of AGI with massive computational resources is an idealized case of AGI, similar to assumptions like the frictionless plane in physics, or the large population size in evolutionary biology. Now that we’ve solved the AGI problem in this simplified special case, we can use the understanding we’ve gained to address more realistic cases. This way of proceeding is mathematically and intellectually rigorous, unlike the more ad hoc approaches typically taken in the field. And we’ve already shown we can scale down our theoretical approaches to handle various specialized problems.

  AGAINST:
  The theoretical achievement of advanced general intelligence using infinitely or unrealistically much computational resources, is a mathematical game which is only minimally relevant to achieving AGI using realistic amounts of resources. In the real world, the simple “trick” of exhaustively searching program space until you find the best program for your purposes, won’t get you very far. Trying to “scale down” from this simple method to something realistic isn’t going to work well, because real-world general intelligence is based on various complex, overlapping architectural mechanisms that just aren’t relevant to the massive-computational-resources situation.

The next section attempts to address the issue of metrics. For any scientific field to be viable, the field must have a means to acquire quantifiable measurements that can be compared and contrasted. Goertzel covers a wide range of proposed approaches including quantifiable, qualitative analysis, and means to measure long term, and incremental term progress towards an ideal AGI. I will address this further in my conclusions.

Conclusions

This review paper was my first introduction to the AGI field. While at first I was a bit disappointed by the vagueness of the field’s direction and purpose, I came to see it as an opportunity to participate in an exciting new field – burgeoning, but adolescent. The whole of AGI is vast, and as of yet has no unified direction or purpose. Therefore acquiring a “forest level” view is difficult and would require studying many different “trees”. What I found most valuable in the paper was the succinct categorization of the fields approaches, giving the reader a decent view of the forest.

As a scientist from a mathematical background I find definitions and metrics to be very important. Goertzel did an excellent job illustrating how difficult it is to consolidate the field of AGI into a succinct definition. Researchers have varying opinions of what the field’s purpose is and what it is they’re working towards. It is only natural then that the field lacks any formalized metrics to measure progress. How can one measure progress if they don’t know what it is they’re working towards? While it is valid to criticize this lack of formalism, there are some who dismiss the field as a “wild goose chase”. Personally, I find this to be an overly harsh censure. Even AGI’s more developed sibling fields such as neuroscience lack the scientific capital to define what general intelligence is.

Recently, some governments have made neuroscience research a larger priority in their budgeting. As a consequence we can hope this increase in scientific vigor will bring us closer to understanding the brain, general intelligence, and as a result, ourselves.

References

1. Goertzel, Ben. “Artificial General Intelligence: Concept, State of the Art, and Future Prospects.” Journal of Artificial General Intelligence 5.1 (2014): 1-46.
2. Kurzweil, Ray. The singularity is near: When humans transcend biology. Penguin, 2005.