Uncertain machines and the opportunity for artificial doubt.
Embracing Uncertainty through AI Models — Words by Pau Garcia
The very instant the alarm clock goes off, we begin to tell our story: “Well, another day, in bed. I’m a hard worker, a good person, someone who always tries their best. Shit. My head hurts, staring at screens late into the night will do that.”
And just like that, through our thoughts, the world comes to exist.
Or, at least, a world exists.
Thinking as a means of creating our reality is an innate, logical, and evolutionary process that is vital for our survival. However, it has its drawbacks. Our thinking is limited to that which is beneficial for us to perceive, just like our sense of sight and hearing. We only see and hear what is necessary for our survival, and we miss out on the rest. We don’t hear the tree falling in the middle of the forest; therefore, it doesn’t fall. Our thoughts are confined to a narrow scope and are geared towards ensuring the well-being of the thinker. Consequently, we are limited in the number of possible eventualities we can consider.
The unfortunate outcome of this limited thinking is the development of the ego. The question of who is striving to survive inevitably leads to the answer — “I”. The mind takes the vastness of the universe and selects one minuscule portion of it, which is the self. From that point of view, it starts narrating, and a sense of reality is born. This notion places us at the center of the universe, and everything that transpires is connected to us in some way. As a result, moral evaluation emerges — actions that benefit us are considered good, whereas those that harm us are deemed bad. The driver that cuts me off in traffic is inherently bad because they are bad for me.
In every waking moment, a delusional space is created between what we imagine and what is really there, and this happens because living is hard. The anxiety of living makes us want to judge, be sure, take a stand, and decide. It is in this strange and unpredictable world that fixed, rigid beliefs can be very comforting for us. That’s why we build up a narrative that splits good people and bad people, encouraging prejudice and magnifying differences.
But in a world full of people who seem to know everything, passionately, based on little information, where certainty is often mistaken for power, what a relief it is to be in the company of someone brave enough to remain unsure.
One of my favorite YouTube videos of all time is an enlightening lecture by the one and only Kurt Vonnegut on the shapes and graphs you could use to map any story, from Hamlet to Cinderella.
In this lecture, you discover certain patterns that have been present in most of the stories and narratives that we have bought into throughout our lives, and I see a lot of similarities between these imposed narratives and the way we use probabilistic data models (AI), these tools just locate the patterns and replicate them with infinite variations.
If only Vonnegut had known he could so easily integrate these storytelling graphs into generative computer systems that would give him back infinite new stories. Interesting as that may be to wonder, the concept that I always return to in this video is the idea of integrated truth in Hamlet. It’s the idea that good stories are the ones that go beyond good or bad, the stories that are able to expose without judgment or moralization, and that give us enough space to be uncertain about what an action, a word, or an ending means.
That is, in Vonnegut’s opinion, (which I support wholeheartedly) the closest to the truth.
In the case of the new conversational models like chatGPT, Bard and so on I find in their answers a lot of certainties, and that is again because these models follow patterns, these models are no more than a mirror to their training database with some complex tweaks in it.
We crave machines that are certain when we are not. We develop technology to help us navigate our anxiety and ignorance, so what use is there for a machine that is uncertain? I truly believe there are many. The worst atrocities we’ve ever committed on ourselves as humans have been the result of single-minded, confident, and intelligent individuals. How might things have been different with a little second-guessing and uncertainty in their decision-making process?
The realm of uncertainty is a space where failure can not only be accepted but praised, where a lack of clarity makes space for improvement.
That’s why our next explorative projects with NLP models will be focused on the building of an Uncertain Machine. A model that helps you question yourself, and understand your ideas and thoughts separately from your inner voice, and without the influence of a general universal voice that most of the AI commercial models amplify.
What do we mean by universal voice? The universal voice is the manifestation of technology that gives you just one answer to any query. That happens with techs like Alexa or Google Assistant and now to ChatGPT, Bard, or any other big conversational language model. An entity that talks like a human to you, and gives answers to most of your questions, but only one, which is, therefore, the best one, the one that is true by default. What happens to all the other options? Who is speaking? Who decides what this voice can answer? What values does this voice follow? To whom does it give more visibility? If we already have problems navigating through Google with all the advertisements, then what will happen if these AI models decide to integrate advertisements as a business revenue source?
We want to build a model that helps us to question ourselves: Am I sure? Is it really so? Is my preexisting opinion causing me to omit something? Is X right or wrong? Is X good or bad? For whom? On what day, and under what conditions? Might there be some unintended consequences associated with X? Some good hidden in the bad that is X? Some bad hidden in the good that is X?
We need machines that give us space for reconsideration. Reconsideration is hard; it takes courage. We have to deny ourselves the comfort of always being the same person, one who arrived at an answer some time ago and has never had any reason to doubt it. A machine that in other words, helps us stay open.
“The test of a first-rate intelligence is the ability to hold two opposed ideas in mind at the same time and still retain the ability to function.”
F. Scott Fitzgerald
The willingness to change your mind is a superpower in the modern age, and I personally believe that we should work with these new technologies to create tools that help us become more open to new ideas, rather than more closed in our own fixed beliefs.
If you want to help us and be part of this journey let us know at: firstname.lastname@example.org
Remix thoughts, influence and words from George Saunders (from the book “A swim in a pond in the rain”, Vonneguts lecture, Alex Johnston edits and myself)