Sci-Fi writers, futurologists and IT researchers and practitioners sometimes conceptualize ‘human-level AI’ as the Holy Grail of AI research. The idea of symbiosis between humans and machines is also settled in mass conscience creating new hopes and new phobias. Will we have a war with machines and end up as their slaves — a slowly-thinking race unable to predict the future and make decisions properly? Or will we live as masters with an army of robotic helpers?
In 2014, a Japanese venture capital company, Knowledge Ventures, elected an AI system to its board of directors. Is it an example of the closest human-machine symbiosis or is it a sign that we are losing our battle with artificial intelligence?
Anthropomorphising intelligence
Since Alan Turing’s times, the major driving force behind AI research has been machine’s competition with human cognition. If we think of such examples as beating humans in chess or simply passing the Turing test — it is either machines proving themselves better than humans or humans outdoing computers in some areas.
This competition is prompted by the fact that the only model we have of anything close to general AI is the human brain. Researchers are inspired by the way our brain is built and how each neuron has thousands of synapses — we can see how it is mimicked in neural networks.
This approach in itself shows the limitations of AI that cannot (and probably will never be able to) fully reconstruct brain functioning. Just as an example, the human brain is very flexible; for instance, it filters information very effectively so that we learn without gigantic amounts of data, whilst AI systems are notorious for their hunger for data.
It goes without saying that Artificial intelligence differs from our brain:
- Artificial intelligence is so far shallow and has limited capacity for transfer.
- It has no natural way to deal with hierarchical structure.
- AI cannot inherently distinguish causation from correlation.
Besides, the whole our world is shaped towards human cognition. We are smart because we are small modules in a big world; we feel part of the society and draw our intelligence and our understanding of the contextual environment from it. AI, on the contrary,
- has not been well integrated with prior knowledge,
- cannot draw open-ended inferences based on real world knowledge, and.
- presumes a largely stable world.
From the psychological side, an essential part of human interaction is empathy and contextual awareness, and we are born with a great intuition for both. It is basically intuition that we are trying to introduce to AI. Currently, instead of brute-forcing its way through the dataset, successful AIs learn to do things by stapling multiple algorithms together. However, machines still fail to generalize much beyond already known data, such as a new pronunciation of a word or an unconventional image, and have trouble dealing with limited amounts of data.
At this point, the most important difference comes into the spotlight: humans have consciousness:
But are so irreparably different? Consciousness is a structure of thoughts, or, at a deeper level, it is just neurons. Consciousness is not binary; it’s a matter of degree. Humans and other animals have different levels of consciousness, and so do adults and children and even different adults. If we stop thinking of machines as continuation of humanity, we can benefit from cooperation with them without feeling threatened by the ghosts of misanthropic robots.
Collaborative intelligence
Lacking consciousness, computers remain task-driven, meaning that they do nothing unless they have a set goal. Humans are those who give the goal and meaning to what AI does for us and with us.
The root idea of collaboration between humans and machines is to enhance each other’s strengths: the leadership, teamwork, creativity, and social skills of the former, and the speed, scalability, and quantitative capabilities of the latter.
This collaboration envisages that every participant has their role, be it a domain specialist getting meaning out of scattered raw data or the selected AI algorithms.
Role of Humans
Training ML algorithms
In many cases, machine learning algorithms are trained with human supervision. Domain specialists collect huge datasets to be fed into algorithms from any field of human knowledge from idioms in multiple languages and disease courses to cultivation of different sorts of apples. Moreover, AI systems undergo training on how to interact with humans to develop just the right personality: confident, caring, and helpful but not bossy. For example, Apple’s Siri was created with the help of human trainers to simulate certain human-like traits.
Connecting AI with the world
As it was mentioned, AI systems have little knowledge of the context — the world surrounding them remains unnoticed and not taken into account. Probably, the most evident example is the emotional deficiency of machines. Humans are driven by emotions. And emotions are precisely the most complex issue to simulate in AI. Consequently, human experts are charged with giving artificial intelligence the correct perception of the factual and emotional surroundings.
Explaining AI behaviour
The famous black-box problem in AI refers to the fact that AI reaches conclusions and renders results through processes that are usually opaque. Since evidence-based industries, such as medicine, a practitioner needs to understand how the AI weighs input data, human experts in the relevant fields are required to explain the machine behavior to users. Such explanations services are becoming integral in regulated industries — the European Union’s General Data Protection Regulation (GDPR), for instance, gives consumers the right to receive an explanation for any algorithm-based decision, such as the rate offer on a credit card or mortgage.
Sustaining AI systems
AI systems should always function properly, safely, and responsibly, that is why they need human supervision to anticipate and prevent any potential harm by AIs. Besides, AI systems should be helped to uphold ethical norms and to protect data privacy.
Role of Machines
When guided by human experts, smart machines may help humans expand their abilities providing fast and well-calculated decisions and insights.
Amplifying human cognitive abilities
AI can boost human analytic and decision-making abilities by providing the right information at the right time. Machines calculate faster and more accurately; they can better categorize or even analyze things, so that human experts receive meaningful preprocessed data.
Interacting with colleagues and customers
AI systems can facilitate communications between people by performing routine tasks, such as by transcribing a meeting and distributing a voice-searchable version to those who couldn’t attend. Such applications are inherently scalable — a single chatbot, for instance, can provide customer service to many people simultaneously.
Embodying human skills
A number of AI-driven applications are embodied in a robot that augments human skills with the help of installed sensors, motors, and actuators. Such robots can now recognize objects and people. They work alongside humans in factories, warehouses, and laboratories to perform repetitive actions that require brute force, while humans carry out complementary tasks where human judgment is needed.
Human-machine cooperation is not always about enhancing our efficiency: it does not require sheer computational power, but relies on intuition, and pre-evolved dispositions toward cooperation, common-sense mechanisms that are difficult to encode in machines. If we could develop the same cooperative disposition in machines — would it be the right degree of consciousness to ensure cooperation?
So far, we see AI mainly as a tool to enhance our physical or cognitive capacities. But what if we find real partners in machines? Machines and humans are a perfect match because they are complementary, and we are here to decide which computer traits we need to develop and use.
References:
Marcus, G. (2018). Deep Learning: A Critical Appraisal. arXiv.
Thiel, P., Masters, B.(2014) Zero to One: Notes on Startups, or How to Build the Future. Currency.
Urban, T. (2015). The AI Revolution: The Road to Superintelligence. Part 1 and Part 2.
