The growth and evolvement of artificial intelligence has showcased the need that AI techniques should be “human-centered”, mainly with the increasing adoption of particularly inscrutable “opaque-box” machine learning (ML) models -such as neural networks models-, in which understanding has become increasingly difficult. For some people, this comprehension challenge will become the bottleneck to trust and adopt AI technologies. Others have warned that a lack of human scrutiny will inevitably lead to failures in usability, reliability, safety, fairness, and other moral crises of AI.
However, what is human-centered artificial intelligence?
HCAI concerns the study of how present and future AI systems will interact with human lives in a mixed society composed of artificial and human agents, and how to keep the human into the focus in this mixed society. The discussion involves both technical and non technical people which see AI agents from different perspectives. HCAI technologies bring superhuman capabilities, augmenting human creativity, while raising human performance and self-efficacy. A human-centered approach will reduce the out-of-control technologies, calm fears of robot-led unemployment, and give users the rewarding sense of mastery and accomplishment. It will as well bring AI wider acceptance and higher impact by providing products and services that serve human needs.
In the past, researchers and developers focused on building AI algorithms and systems, stressing machine autonomy and measuring algorithm performance. The new synthesis gives equal attention to human users and other stakeholders by raising the value of user experience design and by measuring human performance. This new synthesis reflects the growing movement to expand from technology-centered thinking to include human-centered aspirations that highlight societal benefit. The interest in HCAI has grown stronger since the 2017 Montreal Declaration for Responsible Development of AI.
What is the difference between HCAI and AI?
From the standpoint of processes, HCAI builds on user experience design methods of user observation, stakeholder engagement, usability testing, iterative refinement, and continuing evaluation of human performance in use of systems that employ AI.
From the product approach, HCAI systems are designed to be super-tools which amplify, augment, empower, and enhance human performance. They emphasize human control, while embedding high levels of automation.
The goal is to increase human self-efficacy, creativity, responsibility, and social connections while reducing the impact of malicious actors, biased data, and flawed software.
In an automation-enhanced world, clear interfaces could let humans control automation to make the most of people initiative, creativity, and responsibility. For instance, if AI technology developers increase their use of information visualization, their own algorithmic world will improve and they will help many stakeholders to better understand how to use these new technologies. Information visualization has proven its value in understanding deep learning methods, improving algorithms, and reducing errors. Visual user interfaces have become appreciated for providing developers, users, and other stakeholders with a better understanding of and more control over how algorithmic decisions are made for parole requests, hiring, mortgages, and other consequential applications.
Automation is invoked by humans, but they must be able to anticipate what happens because they are responsible. While AI projects are often focussed on replacing humans, HCAI designers favor developing information-rich visualizations and explanations built-in, rather than added on. These information abundant displays give users a clear understanding of what is happening and what they can do.
The future is human-centered. The goal is to create products and services that amplify, augment, empower, and enhance human performance. HCAI systems emphasize human control, while embedding high levels of automation, easing thus the transition from artificial intelligence (AI) to intelligence augmentation (IA).
Although not a recent one (2015), The Master Algorithm by Pedro Domingos is a pleasant book to be read, mainly as a sort of basic pedagogical introduction to machine learning. As the author stated in the book, “when a new technology is as pervasive and game changing as machine learning, it’s not wise to let it remain a black box. Opacity opens the door to error and misuse.” Therefore, this initial effort to democratize this subfield of artificial intelligence is logically welcome.
Professor Domingos is a machine learning practitioner and hence you can realize his bias concerning other approaches to artificial intelligence; said that, it’s interesting how he divides and frames the different schools of thoughts inside machine learning. From his standpoint, there are five schools:
Symbolists: they view learning as inverse deduction and they take ideas from philosophy, psychology, and logic.
Connectionists: they reverse engineer the brain and they are inspired by neuroscience and physics.
Evolutionaries: they simulate evolution on the computer and they draw on genetics and evolutionary biology.
Bayesians: they believe learning is a form of probabilistic inference and they have their roots in statistics.
Analogizers: they learn by extrapolating from similarity judgements and they are influenced by psychology and mathematical optimization.
Each of the five tribes of machine learning has its own master algorithm, a general purpose learner that you can in principle use to discover knowledge from data in any domain. The symbolists’ master algorithm is inverse deduction; the connectionists’ is backpropagation; the evolutionaries’ is genetic programming; the bayesians’ is Bayesian inference; and the analogizers’ is the support vector machine.
For Symbolists, all intelligence can be reduced to manipulating symbols. Symbolists understand that you can’t learn from scratch: you need some initial knowledge to go with the data. Their master algorithm is inverse deduction, which figures out what knowledge is missing in order to make a deduction go through, and then makes it as general as possible.
Symbolist machine learning is an offshoot of the knowledge engineering school of AI. In the 1970s the so-called knowledge-based systems scored some impressive successes and in the 1980s they spread rapidly, but then they died out. The main reason was the infamous knowledge acquisition bottleneck: extracting knowledge from experts and encoding as rules is just too difficult, labor intensive, and failure-prone. Letting the computer automatically learn to, say, diagnose diseases by looking at databases of past patients’ symptoms and the corresponding outcomes turned out to be much easier that endlessly interviewing doctors.
For Connectionists, learning is what the brain does. The brain learns by adjusting the strengths of connections amongst neurons, and the crucial problem is figuring out which connections are to blame for which errors and changing them accordingly. The connectionist master algorithm is backpropagation, which compares a system output with the desired one and then successfully changes the connections in layer after layer of neurons, so as to bring the output closer to what it should be.
Connectionist representations are distributed, mirroring what happens in the human brain. Each concept is represented by many neurons and each neuron participates in representing many different concepts. Neurons that excite one another form a cell assembly. Concepts and memories are represented in the brain by cell assemblies. Each of these can include neurons from different brain regions and overlap with other assemblies.
The first formal model of a neuron was proposed by Warren McCulloch and Walter Pitts in 1943. It looked a lot like the logic gates computers are made of. McCulloch and Pitts’ neuron did not learn though. For that it was needed to give variable weights to the connections amongst neurons in resulting in what’s called perceptrons. Perceptrons were invented in the late 1960s by Frank Rosenblatt, a Cornell psychologist. In a perceptron, a positive weight represents an excitatory connection, and a negative weight an inhibitory one. The perceptron generated a lot of excitement. It was simple, yet it could recognize printed letters and speech sounds, just by being trained with examples.
In 1985 David Ackley, Geoff Hinton, and Terry Sejnowsky replaced the deterministic neurons in Hopfield networks with probabilistic ones. A neural network had then a probability distribution over its states, with higher energy-status being exponentially less likely than lower-energy ones. One year later, 1986, Backpropagation was invented by David Rumelhart, a psychologist at the University of California, with the help of Geoff Hinton and Ronald Williams.
Evolutionaries believe that the mother of all learning is natural selection. The master algorithm is genetic programming, which mates and evolves computer programs in the same way that nature mates and evolves organisms. Whilst backpropagation entertains a single hypothesis at any given time and the hypothesis changes until it settles into a local optimum, genetic algorithms consider an entire population of hypothesis at each step, and these can make big jumps from one generation to the next thanks to crossover. Genetic algorithms are full of random choices; they make no a priori assumptions about the structures they will learn, other than their general form.
Bayesians are concerned above all with uncertainty. The problem then becomes how to deal with noisy, incomplete, and even contradictory information without falling apart. The solution is probabilistic inference and the master algorithm is Bayes’ theorem and its derivates. Bayes theorem is just a simple rule for updating your degree of belief in a hypothesis, when you receive new evidence. If the evidence is consistent with the hypothesis, the probability of the hypothesis goes up, If not, it goes down.
For Analogizers the key to learning is recognizing similarities between situations and thereby inferring other similarities. The analogizers’ master algorithm is the support vector machine, which figures out which experiences to remember and how to combine them to make new predictions. The nearest neighbor algorithm, before the support vector machine, was the first preferred option in the analogy-based learning.
Up to the late 1980s researchers in each tribe mostly believed their own rhetoric, assumed their paradigm was fundamentally better and communicated little with the other schools. Today the rivalry continues but there is much more cross-pollination. For professor Domingos, the best hope of creating a universal learner lies in synthesizing ideas from different paradigms. In fact just a few algorithms are responsible for the great majority of machine learning applications.
As a coda to his pedagogical explanation of machine learning, professor Domingo’s views about data privacy are worthy to be highlighted. From his standpoint, our digital future begins with a realization every time we interact with a computer -whether it’s a smart phone or a server thousands of kilometers away- we do so on two levels: the first one is getting what we want there and then: an answer to a question, a product you want to buy, a new credit card. The second level, in the long run the most important one, is teaching the computer about us. The more we teach it, the better it can serve us -or manipulate us.
Life is a game between us and the learners which surround us. We can refuse to play but then we will have to live a twentieth-century life in the twenty-first. Or we can play to win. What model of us do we want the computer to have? And what data can we give it that will produce that model? Those questions should always be in the back our minds whenever we interact with a learning algorithm -as they are when we interact with other people.
A white paper -published together with Redesigning Financial Services and EY- about how AI can be used to tackle some daily problems the banking institutions have to cope with such as, amongst others, anti-money laundering, KYC, data quality management, process & data mining…
“Yes, but artificial intelligence must become common currency”.
A few days ago I had the good fortune to attend a meeting between technology investors, entrepreneurs, businessmen and professors in the field, the latter three, of AI. It was interesting, on the one hand, to mix in the same virtual space money, willingness to create something, success in having done so, and knowledge… and, on the other hand, to observe the same vital dilemma regarding this technology is shared in the background: democratization or industrialization of artificial intelligence.
Information and communication technologies -and more specifically AI- are GPTs, general purpose technologies, a term coined by MIT professors Erik Brynjolfsson and Andrew McAfee in their book The Second Machine Age; namely, technologies that “disrupt and accelerate the normal march of economic progress”. The steam engine and electricity were also GPTs. They were disruptive technologies that have extended their reach into many corners of the economy and radically altered the way we live and work.
Nonetheless, if we look at the current state of AI, it has yet to take off. Why? One of the reasons is perhaps because it is stuck at a crossroads.
On the one hand, we have tech giants like Amazon, Google, Facebook, Alibaba, Tencent… they are not only competing with each other to see which is the first to discover the next disruptive breakthrough within AI. At the same time, they compete against fast AI startups that want to use machine learning, deep learning, ontologies or even hybrid approaches -mathematics, statistics, rule-based programming and logic…-, to revolutionize certain specific industries. It is a competition between two approaches to extend AI in the field of economics: the industrialization of the powerful giants versus the democratization of the agile startups. How that race plays out will determine the nature of the AI business landscape: monopoly, oligopoly, or free and spontaneous competition amongst thousands of companies. The industrialization approach wants to turn AI into a commodity, with a price tending towards zero. Its goal is to transform the power of AI, and its various subfields, into a standardized freemium service; namely, any company can acquire it, with its use perhaps being free of charge for academic or personal environments. Access to this freemium AI environment would be through cloud platforms. The powerful giants behind these platforms (Google, Alibaba, Amazon…) act as service companies, managing the network and charging a fee. Connecting to that network would allow traditional companies, with a large data set, to leverage the optimization power of AI without having to redo their entire business. The most obvious example of this approach: Google’s TensorFlow. This is an open-source software ecosystem for building deep learning models; however, it still requires specialized programming skills to make it work. The goal of the network approach is to both lower that specialization threshold and increase the functionality of AI platforms in the cloud. Making full use of an AI model is not easy as of today but AI giants hope to simplify this technology and then reap the rewards, in addition to operating the network.
On the other hand, AI start-ups and middle-sized enterprises (MsEs) are taking the opposite approach. Instead of waiting for this network to take shape, they are creating AI niche products for each use case. Such startups and MsEs are aiming at specialization, rather than breadth. Instead of providing, for example, natural language processing models for general purposes, they build new products, solutions, niche platforms for algorithms to perform specific tasks such as fraud tracking, insurance policy comparison, customer profiling for upselling and cross-selling, terrorist threat detection on social networks, pharma knowledge graph generation… The starting postulates of these startups and MsEs are twofold: on the one hand, traditional businesses are still very far, operationally, from being able to use a multipurpose AI network; on the other hand, AI should start to be an intrinsic element in the business operation of these traditional companies. It is because of the latter that, almost always, companies following this approach end up building a strategic relationship with the AI startup or MsE, which has introduced them to this world.
Who will win in this race? Difficult to make a prediction. What is clear is that, if the industrialization approach triumphs, the astronomical economic benefits of this technology will be concentrated in a handful of companies (probably American and Chinese ones); if the democratization approach succeeds, these huge benefits will be spread among thousands of vibrant young agile companies.
As a welcome for the allocated seats in the Redesigning Financial Services Strategic Steering Committee, expert.ai’s Chief Operating Officer Gabriele Donino, and the Managing Director Switzerland Domingo Senise de Gracia were interviewed to talk about the use of artificial intelligence, the potentials, opportunities and barriers.
