Helen Keller Joins the AGI Conversation

The domain of artificial intelligence (AI) is singularly aimed at "recreating full human-like intelligence in machines," says the latest AI reportMichael L. Littman et. al. "Gathering Strength, Gathering Storms: The One Hundred Year Study on Artificial Intelligence (AI100) 2021 Study Panel Report." Stanford University, Stanford, CA, September 2021. by Stanford's professors as they convene every five years to take stock of the past, current, and future directions within AI research and practice. The earlier reportPeter Stone et. al.  "Artificial Intelligence and Life in 2030." One Hundred Year Study on Artificial Intelligence: Report of the 2015-2016 Study Panel, Stanford University, Stanford, CA,  September 2016. pointed to the fact that the more traditional paradigms of AI, including logic-based representation and reasoning, have been displaced and "drowned by the deluge of data and the remarkable success of deep learning."

Speaking of the remarkable success of any deep learning model to date that has truly displaced traditional paradigms of AI, the large language model-based ChatGPT is unrivaled. Lex Fridman posed a very pertinent question to its founder, Sam AltmanPodcast #367, asking whether he believed that a large language model (the one ChatGPT is based on) was really the way to build artificial general intelligence (AGI). Altman replied that this is part of the way and that we need other "super important things." When pressed by Fridman about the missing components, Altman said that any system that can’t go from the knowledge it already has to the stage of discovering or inventing new knowledge isn’t super intelligent. To do that really well, Altman said we'll need to expand on the GPT paradigm in pretty important ways that are still missing.

Amidst the full spectrum of reception to the phenomenon of ChatGPT, which ranges from hype to fear, it’s refreshing to see an honest reflection coming from its founder when it comes to AGI. However, some critical voices from within the computational linguistics and deep-learning community are pointing toward some "super important things" that are missing even from the current set of deep-learning ideas. Professor Yejin Choi, a notable computational linguist, reminds usYejin Choi—“Why AI Is Incredibly Smart and Shockingly Stupid,” TED Talk that predicting the next set of words isn't what human learning is about:

Human learning is never about predicting the next word, but it's about making sense of the world and how the world works.

Professor Gary Marcus, having spent his life intersecting cognitive psychology, neuroscience, and AI, is weary of the shallow learningGary Marcus—“A(G)I: Where we have been, and where we need to go,” 2023 Scottish AI Summit being performed in the name of deep learning by the data-driven models.

To get to a deeper AI that can operate in trustworthy ways even in novel environments, we need to work towards building systems with with deep understanding—not just deep learning.

But where do the folks trained within the narrow bounds of computing go in order to understand how the world works or to find out how we humans deeply understand our reality? What these critical voices are telling their AI community resonates with the future directions pushed by both of the Stanford reports on AI. While in 2016[Over the next fifteen years,] the Study Panel also expects a reemergence of some of the traditional forms of AI as practitioners come to realize the inevitable limitations of purely end-to-end deep learning approaches....We encourage young researchers not to reinvent the wheel, but rather to maintain an awareness of the significant progress in many areas of AI during the first fifty years of the field, and in related fields such as control theory, cognitive science, and psychology. , they advised the AI community to take few domains outside of computing seriously, the latest report's concluding message is louder in terms of inclusivity, lest the technology unintentionally ends up hurting more people, than it helps.

Whereas AI research has traditionally been the purview of computer scientists and researchers studying cognitive processes, it has become clear that all areas of human inquiry, especially the social sciences, need to be included in a broader conversation about the future of the field. Minimizing the negative impacts on society and enhancing the positive requires more than one-shot technological solutions; keeping AI on track for positive outcomes relevant to society requires ongoing engagement and continual attention.

This blog, which brings Helen Keller's whole life and her learning process to the AI research community's table, is one invitation for that broader conversation. This is because Keller and her teacher documented the entire process of her language acquisition. Moreover, Keller has written and spoken much about her journey from learning her first word to having a much more complex understanding of the world, becoming a voice not just for the disabled community but for countless voiceless workers of the Industrial Age. She co-founded the American Civil Liberties Union. She was of the viewHelen Keller. Out of the Dark: Essays, Letters, and Addresses on Physical and Social Vision. Garden City, N.Y.: Doubleday, Page & Co., 1913. that the way to help the blind or anyone with a disability "to understand, correct, remove the incapacities and inequalities of our entire civilization." Every word of hers is an invitation for us to gain an understanding of how the world works.

With the buzz around ChatGPT, the relationship between intelligence and language has been highlighted once more. This blog highlights for the AI community Keller's inspiring relation to the text, and the lifelong learning she embodied. In addition, the AI research community can learn about deep understanding from her learning process. Let's now put ourselves in her shoes as we go through important events in Keller's life.

Due to an illness, Keller lost her ability to see or hear when she was just a few months old. Before her teacher arrived at the scene, Keller was a very active, strong, and playful little girl. Keller could make a couple of people around her understand what she wanted using a self-developed set of hand and bodily gestures, but these signs were not universally understood by those that didn’t know Keller well. While her mother could understand when she needed to drink water and her cook's daughter knew when she wanted to go egg hunting, Keller’s linguistic isolation was becoming increasingly frustratingMeanwhile the desire to express myself grew. The few signs I used became less and less adequate, and my failures to make myself understood were invariably followed by outbursts of passion. I felt as if invisible hands were holding me, and I made frantic efforts to free myself. as she grew.

The most important day I remember in all my life is the one on which my teacher, Anne Mansfield Sullivan, came to me. I am filled with wonder when I consider the immeasurable contrast between the two lives which it connects. It was the third of March, 1887, three months before I was seven years old.

We must put ourselves in Helen Keller's shoes to be able to empathize with her world, a world that is without sight or sound. Keller’s combination of blindness and deafness meant that in her early childhood, Keller did not acquire language in the same manner as many children. Sullivan believed that Keller was perfectly capable of learning; she just needed to be taught in an accessible way. In Keller's own words from her autobiographyHelen Keller—The Story of My Life:

The morning after my teacher came she led me into her room and gave me a doll. The little blind children at the Perkins Institution had sent it and Laura Bridgman had dressed it; but I did not know this until afterward. When I had played with it a little while, Miss Sullivan slowly spelled into my hand the word “d-o-l-l.” I was at once interested in this finger play and tried to imitate it. When I finally succeeded in making the letters correctly I was flushed with childish pleasure and pride. Running downstairs to my mother I held up my hand and made the letters for doll. I did not know that I was spelling a word or even that words existed; I was simply making my fingers go in monkey-like imitation.

To be able to imitate already meant that Keller possessed a sense (at least of touch) and a memory. Using the deaf-blind manual alphabet, an alphabet based on sign language, Miss Sullivan was able to spell words into Keller’s hands, no doubt. But how could Keller learn to identify what those words were and what this whole exercise meant? For Keller, this kind of learning also required repetition.

One day, while I was playing with my new doll, Miss Sullivan put my big rag doll into my lap also, spelled “d-o-l-l” and tried to make me understand that “d-o-l-l” applied to both. Earlier in the day we had had a tussle over the words “m-u-g” and “w-a-t-e-r.” Miss Sullivan had tried to impress it upon me that “m-u-g” is mug and that “w-a-t-e-r” is water, but I persisted in confounding the two. In despair she had dropped the subject for the time, only to renew it at the first opportunity.

Sullivan was right. Initially, every human learner is merely mimicking without really knowing what they are doing, and it takes a lot of learning instances during the learning process to reach a point of comprehension. But what is all the training process aimed at? To be able to know!

We walked down the path to the well-house, attracted by the fragrance of the honeysuckle with which it was covered. Some one was drawing water and my teacher placed my hand under the spout. As the cool stream gushed over one hand she spelled into the other the word water, first slowly, then rapidly. I stood still, my whole attention fixed upon the motions of her fingers. Suddenly I felt a misty consciousness as of something forgotten—a thrill of returning thought; and somehow the mystery of language was revealed to me. I knew then that “w-a-t-e-r” meant the wonderful cool something that was flowing over my hand. That living word awakened my soul, gave it light, hope, joy, set it free! There were barriers still, it is true, but barriers that could in time be swept away.

Keller could, of course, touch things. Like every child, Keller had her favorite toys. After Sullivan’s lessons, she could also mimic hand gestures or spellings. The only missing cognitive connection was that she didn't know that the things she touched and the sequence of hand gestures that she'd been made to do were symbolically connected—that the thing had a name and that the name stood for the thing.

To be able to name things might sound trivial, but doing so is the core of language and communication, and this cognitive ability is what all human knowledge and understanding stand upon. Learning begins after this realization. Keller says:

I left the well-house eager to learn. Everything had a name, and each name gave birth to a new thought. As we returned to the house every object which I touched seemed to quiver with life. That was because I saw everything with the strange, new sight that had come to me.

Helen Keller went on to eventually become an agent of change, and an agent of change does not just describe the world as it is sorrily functioning at the time; an agent of change also creates a better world. Through Keller's example, we come to understandHelen Keller—The World I Live In that the change has to first take place cognitively, through words.

Language, ... is not primarily a way of describing or referring to the world. It is, rather, our means of creating, of shaping, of populating the world; it is what makes the world our world. Before my teacher came to me, I did not know that I am.

This blog is especially interested in exploring the learning process through which Keller acquires the general intelligence that goes from uttering a simple noun phrase—a mere description of the world ("this is water") to embodying her own advice to the a young woman who wrote to her, that the way to help the blind or any community of disability is "to understand, correct, remove the incapacities and inequalities of our entire civilization."

With the help of Keller’s teacher Anne Sullivan's correspondence through letters with her own teacher, we can peek into some of the initial important stages of Keller's learning as she went from single words to sentences, as she moved from nouns to adjectives to prepositions, as she learned emotive words, as she learned to reason, and as she learned to read classic poetry and books that eventually shaped her understanding of her own thoughts and the place she occupied in this world.

Initially, like any child, Keller communicated whole sentences using only single words or gestures, but through the repeated insistence of her teacher on complete sentences, she gradually internalized the idea and began to use more complete sentences herself. Keller then learned adjectives and adverbs just like she learned nouns, grasping the concept before associating it with the word. Since she already had signs for "smalltake up a tiny bit of the skin of one hand between the thumb and finger of the other" and "largeshe spread the fingers of both hands as wide as she could, and brought them together, as if to clasp a big ball" before being introduced to the corresponding words, the teacher only had to substitute the words small and large for these signs. The same capacity to experience the sensations is what helped Keller in learning the linguistic counterparts for different states of feeling and emotions.

Every child interprets the world as well as each word according to their worldview—the same was the case with Keller. This worldview is ever-evolving, of course. Teacher Anne Sullivan ensured that she took Keller outside in nature to understand the world and inside the world of books to understand the word. These trips helped build the experiences as well as vocabulary simultaneously by feeding Keller's imagination and worldview. Another thing working in favor of Keller is her impulse to share her new findings with her mother. Sullivan narrates:

She makes many mistakes, of course, twists words and phrases, puts the cart before the horse, and gets herself into hopeless tangles of nouns and verbs; but so does the hearing child. I am sure these difficulties will take care of themselves. The impulse to tell is the important thing. I supply a word here and there, sometimes a sentence, and suggest something which she has omitted or forgotten. Thus her vocabulary grows apace, and the new words germinate and bring forth new ideas; and they are the stuff out of which heaven and earth are made.

Eventually, Keller started reading books, and Miss Sullivan ensured that these were the same classic books that any child growing up in the world would be reading. But they made a game out of reading—finding words that Keller already knew. Sullivan says that they competed to see who found the words faster.

She is as triumphant over the conquest of a sentence as a general who has captured the enemy’s stronghold.....When asked [much later] why she loved books so much, she once replied: “Because they tell me so much that is interesting about things I cannot see, and they are never tired or troubled like people. They tell me over and over what I want to know.

How beautiful it would be to slowly but surely be able to have a meaningful conversation, and it would be equally beautiful and rewarding a feat for both Sullivan as well as for Keller:

When the sun got round to the window where she was sitting with her book, she got up impatiently and shut the window. But when the sun came in just the same, she came over to me with a grieved look and spelled emphatically: “Sun is bad boy. Sun must go to bed.”

She is the dearest, cutest little thing now, and so loving! One day, when I wanted her to bring me some water, she said: “Legs very tired. Legs cry much.”

It wasn't long before the experiences and their innocent expressions eventually turned into reflections and inquiries. It is through Keller that we understand the importance of the simple words that give birth to such inquiries of the mind:

She has now reached the question stage of her development. It is “what?” “why?” “when?” especially “why?” all day long, and as her intelligence grows her inquiries become more insistent. I remember how unbearable I used to find the inquisitiveness of my friends’ children; but I know now that these questions indicate the child’s growing interest in the cause of things. The “why?” is the door through which he enters the world of reason and reflection. “How does carpenter know to build house?” “Who put chickens in eggs?”

The following excerpt from one of Sullivan's letters summarizes the entire evolutionary process of a mind's education and how the growth of one's understanding of the world and one's vocabulary go hand in hand:

She continues to make rapid progress in the acquisition of language as her experiences increase. While these were few and elementary, her vocabulary was necessarily limited; but, as she learns more of the world about her, her judgment grows more accurate, her reasoning powers grow stronger, more active, and subtle, and the language by which she expresses this intellectual activity gains in fluency and logic.....As her observation of phenomena became more extensive and her vocabulary richer and more subtle, enabling her to express her own conceptions and ideas clearly, and also to comprehend the thoughts and experiences of others....Her mind should begin to put forth its higher powers, and generalize from innumerable impressions and ideas which streamed in upon it from books and from her daily experiences. Her mind sought for the cause of things.

Without intending to trivialize the entire phenomenon, it’s worth reiterating that there is a cognitive chasm between the object in the world and the symbol that stands for the same object. That chasm is bridgeable, but, as made evident by Keller’s story, the bridge has to be crossed by the learner. Eventually, Helen Keller was able to bridge this chasm, as she clearly went from knowing one word to a very subtle and complex understanding of how the world and society work—a verifiable fact found in her written expression in published books. Keller grew up in the Industrial Age, and she became a voice for the workers of that era. Even though some non-disabled folks would rather have had her just work for the betterment of the disabled community, Keller stayed true to her inclusive ethos. Today, her voice would surely have been a part of the Stanford professors' table while the future of the AI age is being shaped.

These were all anecdotes, no doubt autobiographical and personal and very educative for the AI community, but perhaps there's a theory of meaning-making that can help explain to Sam Altman and the rest of the AI community how deep learning generally takes place—learning aimed towards a deeper understanding of the way the world works. Nevertheless, it seems clear that the founding father of AI, Alan Turing, was right in suggestingInstead of trying to produce a programme to simulate the adult mind, why not rather try to produce one which simulates the child’s? If this were then subjected to an appropriate course of education one would obtain the adult brain. [A. M. Turing (1950) Computing Machinery and Intelligence. Mind 49: 433-460.] that a child's mind can be studied as a model of intelligence for the AI research community, and Keller herself is somewhat unique in that she has written quite a bit about how she remembers acquiring language as a child.

Is AI close to achieving its goal of computing artificial general intelligence? Does ChatGPT know what the words it is generating stand for and mean? Clearly not! But at the moment, it can mimic well while replacing words with similar words without ever grounding them in reality. That's a start! Nevertheless, to find out how it is seamlessly able to pay attention to the intention of its user's prompts, we can learn how the state-of-the-art, attention-based transformers have improved upon the earlier sequence-to-sequence models in this wonderful course by Educative: