tHe FeAr oF AI?
Before we start we need to lay out the foundation with regard to general terms and concepts.
Artificial intelligence is the broadest term used to describe the approach toward problem-solving using computer science, machine learning is a subset of AI, representing algorithms used to build a model based on sample data(training data), in order to make predictions or decisions without being explicitly programmed to do so (e.g. linear regression), whereas deep learning is a subset of machine learning, with the advantage that it eliminates some of the toils of feature extraction (reducing the number of resources required to describe a large set of data).
A brief overview of types of machine learning: supervised learning (uses labeled datasets), unsupervised learning (uses unlabeled datasets), and reinforcement learning (the model learns through reward maximization aka. feedback ). Deep learning can leverage labeled datasets, but it can also ingest unstructured data /unlabeled datasets (e.g. images) to determine a set of features.
In order to reach your destination you have to know the starting point…let’s quickly recap some of the recent breakthroughs in the field of AI with their associated papers:
- DeepFace : Facebook’s face recognition system with an accuracy of 97%.
- AlphGo : DeepMind’s deep learning neural network managed to defeat legendary Go player Lee Sedol at the game of Go.
- AlphaFold : DeepMind’s model is capable of predicting 3D models of protein structures and has the potential to accelerate research in every field of biology by solving the problem of protein folding, and was placed first in the overall rankings of the 13th Critical Assessment of Techniques for Protein Structure Prediction (CASP).
- GPT-3 : OpenAIs autoregressive language model (probability distribution over sequences of words) that uses deep learning to produce human-like text.
So why do people tend to anthropomorphize AI…It’s not so surprising is human nature, we all see people who name their cars. Of those aforementioned breakthroughs, which one do you think had to most impact in terms of social exposure?
If you said GPT3, well you guessed it…and now (June 2022) another AI-related story is making some waves in the mainstream media and that’s the Is LaMDA Sentient story.
But why? Simply because it’s easy to misinterpret embodiment for agency, and also it’s somehow intellectually accessible, basically it’s a cookie that our brain bakes for itself.
What does it mean? Let’s say you record your voice and put that recording into an object (e.g toy) a naive observer might say that particular toy is “alive”, but that’s an embodiment case (an “agent” that inhabits a subject other than itself), on the other hand, human agency is much more, it represents the ability to act and to demonstrate causal reasoning, and the current AI systems cannot do this, although they can mimic it pretty well.
Lately, we’re witnessing the emergence of text-to-image AI systems in the creative domain, like OpenAI’s DALL.E and Google Imagen, which are artificial neural networks with the ability to generate photorealistic images from brief yet descriptive sentences.
Which is quite impressive, and the output pictures have unprecedented photorealism, at a first glance it seems that these AI systems are mainly fun but they can have very practical uses (also as deepfake, depending on which side of the table you decide to be…” when seeing is no longer believing ”).
In conclusion, I would say despite major recent advances, AI is nowhere close to even matching the breadth and depth of perception, reasoning, communication, and creativity of people.