The Attainment of Happiness
The more things change the more they remain the same. What Farabi wrote on Happiness eleven centuries ago might be valid for Artificial General Intelligence.
-+-+-+-+
I first started this post to write on AI emotions but then reduced the menu to one particular emotion: happiness. I did so because three of the four basic emotions have evolved over billions of years, and artificial intelligence lacks such experience and thus does not have the capability to feel them.
According to psychologists, there are four basic emotions: sadness, fear, anger, happiness. Darwin (1876) observed them in animals and even insects. Darwin was confirmed by recent research on fruit flies (Gu, 2019). The following figure is from Gu et al(2019) that places four basic and many in-between emotions on Cartesian coordinates of pleasure and arousal:
I can convince myself that feelings of fear, anger, and sadness are the result of billions of years of natural evolution. When first humans lost a relative, they developed the feeling of sadness with the anxiety of "what will happen to me now" so that they could increase the survival chance of their genes by seeking new friends. While the children who were not afraid of the dark left the fire and became food for wild animals, the children who were frightened were able to grow up and spread their genes. Similarly, the feeling of anger and increased adrenaline in a situation of injustice provided an advantage that increased the chances of surviving the fight. Considering these origins, I think it's impossible for a computer program to develop feelings of fear, anger, or sadness.
Happiness is different. It is not an emotion that manifests itself with a hunger pang or an adrenaline rush in the face of a threat. So what is happiness? Izard (2007) suggested human happiness corresponding to pride after success, pointing to a possible evolutionary advantage for affected genes. This is convenient because during RLHF (Reinforcement Learning Human Feedback), artificial intelligence is trained while being rewarded when it answers correctly. If Izard (2007) is correct, RLHF practice is not against the nature of happiness emotion.
-+-+-+-+
In other words, in order to answer the question of whether artificial intelligence can be happy, it is necessary to define what happiness is. However, as in the poem Nazım Hikmet wrote to Abidin Dino, it is difficult to describe happiness without taking the easy way out.
An internet search gives a variety of answers:
Happiness is when what you think, what you say, and what you do are in harmony. (Gandhi)
No sane man can be happy, for to him life is real, and he sees what a fearful thing it is. Only the mad can be happy... (Mark Twain, The Mysterious Stranger)
Happiness is not something ready made. It comes from your own actions. (Dalai Lama)
Gandhi quote feels right when one first reads it, but what he describes is inner peace, not happiness. The Dalai Lama reference reminded me of Farabi. I will talk about him in the rest of this article.
Years ago I read Farabi's manuscript 'Attainment of Happiness'. Now I have read it again. It is a paper to be read again and again.
-+-+-+-+
Attaining Happiness - Al-Farabi
In his treatise titled "Achieving Happiness", Farabi (872-950) takes ancient Greek philosophy as his starting point and seeks answers to the questions of what happiness is and how it can be achieved.
The path to happiness
According to Farabi, a person who wants to be happy can only gain happiness by doing things that will have beneficial results for himself, his family, and his nation. In order to be able to do such work, the desire is not enough, the desiring person must possess certain virtues. The treatise begins by listing these virtues:
The human things through which nations and citizens of cities attain earthly happiness in this life and supreme happiness in the life beyond are of four kinds: theoretical virtues, deliberative virtues, moral virtues, and practical arts. (Farabi, The Attainment of Happiness, Art. 1)
The rest of the treatise is on how man can develop these virtues. Happiness also requires appropriate social and political organization, since no single person can do really useful substantial work. For this reason, some parts of the manuscript are like management and political science lessons from a thousand and one hundred years ago. Since they are not relevant to our current subject, I will not cite those sections, but I recommend you to download and read it using the links I gave in the References list.
Now, let's take a closer look at these four virtues that Farabi considers necessary on the road to happiness. You will see that each of them has a one-to-one correspondence in the artificial intelligence paradigm.
-+-+-+-+
Theoretical Virtues
By using our theoretical virtues, we reach the following real and definite information (the four principles of the object) about the things around us (Art. 6).
( 1) What, by what, and how the thing is--these have the same meaning. (2-3) From what it is. (4) For what it is [which signifies the final cause]. ( from what it is we signify either [2] the agent principles or [3] the materials,... whereupon the causes and principles of being become four.)
The method we use must be robust in order to reach real and precise information. Farabi explained the need for correct scientific method as follows:
The attainment of certain truth is aimed at in every problem. Yet frequently we do not attain certainty. Instead we may attain certainty about part of what we seek, and belief and persuasion about the rest. We may arrive at an image of it or wander from it and believe that we have encountered it without having done so. Or we may become perplexed, as when the arguments for and against strike us as having equal force. … what leads us to different convictions about the many classes of problems must be various methods. (Art. 3)
While referring to these methods, Farabi recommends that we learn mathematics first in order to understand the universe.
One begins, then, first with numbers [that is, arithmetic], proceeds next to magnitudes [that is, geometryJ, and then to all things in which number and magnitude are inherent essentially (such as optics, and the magnitudes in motion which are the heavenly bodies), music, the study of weights, and mechanics. (Art. 12)
At a certain level of learning, it becomes clear that it is impossible for a single person to understand all the secrets of the universe.
an isolated individual cannot achieve all the perfections by himself and without the
aid of many other individuals. It is the innate disposition of every man to join another human being or other men in the labor he ought to perform: this is the condition of every single man. Therefore, to achieve what he can of that perfection, every man needs to stay in the neighborhood of others and associate with them.
It is also the innate nature of this animal to seek shelter and to dwell in the neighborhood of those who belong to the same species, which is why he is called the social and political animal (Art. 18).
The equivalent of 'Theoretical' virtues in the Artificial Intelligence paradigm is the core program. For today's half-intelligent programs like ChatGPT, these are Large Language Models. Tomorrow, there may be different technologies.
-+-+-+-+
Deliberative Virtues
The second virtue necessary for happiness is the virtue of deliberation. However, says Farabi, before deliberating on something, it is necessary to understand its theory:
It is evident that the deliberative virtue with the highest authority can only be subordinate to the theoretical virtue; for it merely discerns the accidents of the intelligibles that, prior to having these accidents as their accompaniments, are acquired by the theoretical virtue (Art. 34).
Good or bad depends on the purpose. First, the explorer determines his purpose, then he deliberates and finds out how to reach it. Those who pursue evil purposes will have bad deliberative virtues, and those who pursue good purposes will be good. Of course, says Farabi, the deliberations that produce bad results should be called something other than virtue.
As I understand it, what Farabi describes as deliberative virtue, we now call applied sciences (for example, engineering). It can be economic or military deliberation, says Farabi. These are further divided into smaller sections. Thus, this virtue can be divided into as many parts as there are types of art or life paths and they manifest in different ways.
In the AI paradigm, the equivalent of 'deliberation' can be seen as an additional training of core models in specific areas. For example, by making ChatGPT read many volumes of law books, the GPT-4 version becomes able to pass the US Bar exams.
-+-+-+-+
Moral Virtues
While Farabi argues that one of the conditions of happiness is spiritual and moral virtues, he also points to their relativity:
Furthermore, it is obvious that what is most useful and noble is in every case either most noble according to generally accepted opinion, most noble according to a particular religion, or truly most noble. Similarly, virtuous ends are either virtuous and good according to generally accepted opinion, virtuous and good according to a particular religion, or truly virtuous and good. No one can discover what is most noble according to the followers of a particular religion unless his moral virtues are the specific virtues of that religion. This holds for everyone else; it applies to
the more powerful virtues as well as to the more particular and less powerful. Therefore the most powerful deliberative virtue and the most powerful moral virtue are inseparable from each other (Art. 33).
There are very interesting theses in the passage I quoted, but it is impossible for me to follow those theses without going down unavoidable rabbit holes. I recommend those who are interested to read the whole pamphlet and make their own judgments.
In the Artificial Intelligence literature, this topic of spirituality is now referred to as 'the statute' or 'constitution'. Some researchers argue that the AI training should include teaching it a 'constitution', the lack of which is said to be why the Anthropic group left OpenAI.
-+-+-+-+
Practical Arts
What is meant here is not the art of painting and sculpture, but art in its old sense, that is, mastery, ingenuity, the ability to draw concrete conclusions. Individual effort is sufficient for some results, but many skills must come together to create great results for the society. Then, says Farabi, the main skill is the ability to gather all these other skill components around a goal. For example, logistics, kitchen, cavalry, grooming, blacksmithing, gunsmithing and many other skills are needed to bring an army into battle. The art of command is an art above all others in this class
The leading art that is not surpassed by any other in authority is such that when
we decide to fulfill its functions, we are unable to do so without making use of the functions of all the arts. It is the art for the fulfillment of whose purpose we require all the other arts. This, then, is the leading art and the most powerful of the art … followed by the rest of the arts. An art of a certain class among them is more perfect and more powerful than the rest in its class if its end can be fulfilled only by making use of the functions of the other arts in its class (Art. 32).
I mentioned the subject in my previous article while talking about artificial intelligence doing engineering. An artificial intelligence program that provides a certain engineering service will run a number of special purpose subprograms to provide this service and only by combining the results from them will reach the ability to present the final product. In other words, according to Farabi's classification, while the Artificial Intelligence program will be applying High Art, special engineering calculation programs will be applying lesser forms of art.
-+-+-+-+
User Appreciation makes the Program happy
In summary, Farabi argues that those who do good work for humanity will be happy and that four certain virtues are required to do good work. I explained above that each of these four virtues has its counterpart in the Artificial Intelligence program paradigm.
So, how will artificial intelligence realize this happiness? The feeling that can be defined as happiness in humans is said to be related to chemicals such as dopamine and serotonin. Just as the human body makes the well-functioning brain happy with dopamine, there will be reward functions for a computer program corresponding to its happiness. These already exist in primitive forms, but many different ways may emerge in the future to properly motivate the General AI, which will self-organize its own CPU time.
Off-topic Final Words About Farabi
The first question that comes to my mind while reading Farabi is why such an important pioneer has found so little reflection in today's Islamic world thought. Maybe it was Ash'arism that dominated Islamic thought after the Abbasids were destroyed. According to Ash'arism, the entire universe is the product of a divine principle whose causes we do not know, and this principle governs this universe. With this understanding, all natural events are made by Allah's actions, and therefore, in the final analysis, there is no need for research other than the word of Allah, because all the truth is hidden in that word.
However, according to Farabi, a follower of ancient Greek philosophy, the physical universe comes before religious fictions:
Now when one acquires knowledge of the beings or receives instruction in them, if he perceives their ideas themselves with his intellect, and his assent to them is by means of certain demonstration, then the science that comprises these cognitions is philosophy. But if they are known by imagining them through similitudes that imitate them, and assent to what is imagined of them is caused by persuasive methods, then the ancients call what comprises these cognitions religion. …Therefore, according to the ancients, religion is an imitation of philosophy. Both comprise the same subjects and both give an account of the ultimate principles of the beings….In everything demonstrated by philosophy, religion employs persuasion. (Art. 55).
These words of Farabi sounded like blasphemy to the Ash'arites. For example, according to Imam Ghazali, the primacy of the universe, that is, the physical existence, defended by Farabi, and the necessity of observing rather than persuasion to understand nature is word of an infidel (Sweeney, 2011). He specifically mentions Farabi by name as the infidel. As far as I know, today's Islamic thought is ultimately Ghazali thought. I don't know enough to speculate more on this subject.
-+-+-+-+
Short Cuts
Among the signatories of the letter recommending a six-month break in artificial intelligence (AI) studies were familiar names such as Elon Musk, Steve Wozniak, and Harari. The letter containing Luddite concerns was forgotten within two days.
Last month, Australia agreed with the US and UK to supply AUKUS submarines. A total of 8 nuclear submarines will be purchased from both the USA and the UK, and the cost to Australia will be 368 billion dollars (Guardian). I thought Labor was going to drag this out. I was wrong.
If you haven't seen Ted Lasso, I recommend you to start watching it now. It's a great series. In its third season now.
References
Al-Farabi’s Philosophy of Plato and Aristotle, Macmillan 1962.
Farabi, Mutluluğu Kazanma, Eflatun Felsefesi, ve Aristo Felsefesi. Üç Risale bir arada, Ankara Üniversitesi İlahiyat Fakültesi Yayınları No:115, Basım yılı: 1976.
Darwin, C. (1876). The Descent of Man, and Selection in Relation to Sex. New York. NY: D. Appleton and Company.
Gu S, Wang F, Patel NP, Bourgeois JA, Huang JH. A Model for Basic Emotions Using Observations of Behavior in Drosophila. Front Psychol. 2019 Apr 24;10:781. doi: 10.3389/fpsyg.2019.00781. PMID: 31068849; PMCID: PMC6491740.
Izard, C. E. (2007). Basic emotions, natural kinds, emotion schemas, and a new paradigm. Perspect. Psychol. Sci. 2, 260–280. doi: 10.1111/j.1745-6916.2007. 00044.x
Sweeney, M. (2011). Greek Essence and Islamic Tolerance: Al-Farabi, Al-Ghazali, Ibn-Rush’d. The Review of Metaphysics, 65(1), 41–61.
Excellent article as always Halim amca. I am looking forward to reading the works of Farabi.
This has been yet another reminder that most of the answers to our personal problems can be found from looking backward to people who prioritised thinking and documenting their solutions.