George Krasadakis
Artificial Intelligence. One of the most popular technology terms of our time— and very frequently, overused or even misused.
The media loves both the success stories and ‘dystopias’ driven by Artificial Intelligence. Machines replacing human workers, AI exceeding human intelligence, robots taking control and so on.
If you look beyond this hype, you will realize that there is a real revolution happening. To understand the potential of AI, just examine the recent advances in fields like Deep Learning and their applications in domains such as Computer Vision and Natural Language Processing.
There is a massive disruption in progress — powered by a combination of technologies, enabling machines to make sense of massive volumes of data and perform cognitive functions.
AI is changing our world and the impact to come is massive: on the way we work, we live, collaborate, decide and act as a society.
1. Artificial Intelligence, defined
Artificial Intelligence can be defined as the technology enabling systems to encapsulate cognitive functions along with adaptive and learning capabilities — leading to self-improvement.
AI-powered systems can capture and ‘understand’ their environment and make optimal, real-time decisions towards specific objectives.
As a characteristic example of AI, ‘Computer Vision’ enables systems to ‘see’ via sophisticated algorithms. These are trained to identify a wide range of entities such as landscapes, persons and objects in a picture or video.
In another example of applied AI, ‘Natural Language Processing’ technologies enable interaction with a machine based on free-form, natural, speech.
NLP and related technologies can ‘understand’ natural speech and respond in a meaningful way. As soon as the machine extracts the context of the ‘natural speech’ request, it synthesizes the right response which is also served back to the user as ‘natural speech’.
The rapid progress of AI is empowered by streams of data on major human activities. These include online communication, social interaction, device usage, searches, content consumption and IoT data streams — to name only a few.
To make sense of these vast amounts of complex data, AI systems leverage the power of cloud computing and specialized machine learning algorithms. World-scale data centres, with huge, labelled data sets are being used for training AI algorithms in performing certain cognitive functions.
2. The state of the art
Algorithms can now ‘see’
The ability for a computer to ‘see’ is an astonishing achievement. AI-powered systems can ‘understand’ the context of an image or a video in impressive level of detail. They can identify an expanding set of entities — such as persons, named individuals, cars, houses, streets, trees and more — with increasing levels of success.
Given an image or video, algorithms can estimate additional properties such as the number of persons in the picture, their gender, age or even their emotional state.
You can simply submit a family photo to one of the commercially available cognitive services, and get in milliseconds a response with the persons identified, their gender, age and the dominant emotions. An object in a photo can also be identified. For example, AI can recognize a car and also its maker and model. It can then tag it for improved searching, grouping and discoverability.
In the close future, algorithms will be able to infer even the situation implied — such as a kids party, a sports event, a business conference or a random arrangement of people in a park.
The possible applications of computer vision are impressive. From autonomous cars which can ‘see’ in 360 and understand their environment and its dynamics in real-time, to special applications like the Seeing AI by Microsoft — a prototype system helping people who are visually impaired or blind to understand their environment!
Computer vision is making huge steps, with massive applications in autonomous cars, navigation, robotics, pattern recognition, medical diagnosis and more. AI systems keep learning, and they learn fast.
The dialogue with the ‘machine’
A short interaction with Amazon Alexa, Cortana, Siri or Google Assistant is sufficient to realize the huge progress of Natural Language Processing technologies.
Microsoft and IBM announced that their NLP technologies perform at the same level (or better) compared to professional transcribers in processing discussions ranging from sports to politics
Google recently demonstrated Duplex, its digital assistant technology, which is able to complete certain tasks via a natural conversational experience. For example, it can arrange a meeting or appointment via a free-form dialogue with a human.
Digital assistants will become more and more intelligent, contextual and proactive.
At some point in the not so distant future, your digital assistant will respond naturally, in a conversational mode and possibly with a style, attitude and humour matching your personality and your current mood.
Digital assistants continuously learn using each single interaction with the user. They better match the user’s explicitly stated or implicitly identified preferences. At some point in time, DAs will become proactive and autonomous by seamlessly leveraging deep knowledge about the user, signals from user’s environment and global trends and dynamics.
3. Applications and Impact
AI is already impacting our socioeconomic system in many ways. We have entered a phase of drastic transformation of markets, businesses, education, government, social welfare systems, companies, employment models and social structures. All will all be soon re-shaped as the result of intelligent technologies and automation. The massive adoption of AI will fundamentally change all industries, as summarized below.
Transportation is already in a transformation mode. Fully autonomous cars will soon be a reality. They will be safer, more efficient and more effective. Autonomous trucks, smart containers, driverless taxis and smart cities are just some examples of the reality to come for the transportation industry.
AI in transportation will drive massive changes, not only to the vehicles, but also to the entire ecosystem — from taxi services to e-commerce and package delivery services.
Consumer habits will be severely impacted, with a shift from owning a car to consuming car services on demand.
The cost of a vehicle as a service will be significantly lower due to, among other factors, the capability of better utilization of the cars by the company operating the service.
Entire transportation networks consisting of fleets of autonomous cars will be orchestrated by AI algorithms to best adapt in real-time, to demand, traffic and other conditions. This will transform the way people commute along with the way cities expand and grow.
For example, the new era of cheaper, faster and safer transportation with autonomous vehicles, might trigger a de-urbanization trend — especially if you consider that the time spent in autonomous vehicles can be fully productive with the capabilities of a modern office.
Electronic commerce. Customer experience is becoming smarter with advanced, AI-powered personalization, dynamic pricing and offer generation.
Fulfillment centers become more automated — with robots navigating the space to collect products and execute customer orders — in some cases, autonomously.
Driverless drones and cars could have a role in the last part of the delivery process. As centralized intelligence will orchestrate the entire processes, typical sales processes, channels, networks of physical stores are becoming less important — thus disrupting the industry.
Financial services, insurance and any other sector requiring significant amount of data processing and content handling will also benefit from AI.
Financial institutions will automate significant processes regarding transaction validation, fraud identification, stock trading, recommendation and advisory services.
Insurance companies will leverage the vast amounts of data available and predictive and machine learning technologies, to get better risk estimations. As a result, they will be in position to offer better products, matching the exact needs of a certain customer. Car insurance companies will also be significantly impacted by the adoption of smart, driver-less cars.
The state and citizen services. AI can have a great impact in eliminating bureaucracy, improving citizen services, governance and social programmes.
Legal services. Even more traditional professions which are built on top of strong relationships, such as legal professions, will be re-defined by AI. Typical support services in a legal context, deal with document handling, classification, discovery, summarization, comparison and knowledge management — tasks where AI agents already excel.
Product development. AI introduces new capabilities changing the typical product development process — for digital or physical products. With the general availability of advanced cognitive technologies (cloud-based commercial AI offerings via easy-to-consume APIs) and the low-cost integration scenarios, the AI-powered opportunities for innovation increase exponentially.
Commercial cognitive APIs and the cloud make it easy for software developers to build cognitive apps, powered by advanced AI capabilities. Physical product manufacturing processes can also benefit from AI-powered production lines, quality control systems and continuous improvement processes. Products will soon be built in totally different ways; and they will be connected and intelligent.
Education. The overall education system will be dramatically improved by AI on top of world-scale digitized content, data and scientific and general knowledge.
Intelligent education agents will be capturing the needs of the student to synthesize optimal personalized educational programs — matching the intent of the student, the right level, pace, preferred types of content and other parameters.
In another scenario, AI-powered apps will be able to recommend education opportunities and personalized educational content, proactively — depending on the current state of user’s career, education level and previous experiences.
This could take the form of an always-on, intelligent ‘education adviser’, discovering the right learning opportunities for each user.
4. The concerns
There are serious concerns and unanswered questions regarding the social, political and ethical implications of massive adoption of AI.
For instance, the ‘intelligent automation’ which can be achieved at scale by using Artificial Intelligence, is expected to transform the way we work and the skills in demand. Certain roles will become obsolete and some professions will eventually disappear.
Lethal Autonomous Weapons. The concept of an autonomous machine is impressive. Think of an autonomous car, which can capture its environment and dynamics and make real-time decisions, to achieve a predefined objective — move from point A to B — under certain constraints.
In a military context though, this autonomy in decision-making is frightening: the so called Lethal Autonomous Weapons, refer to futuristic robotic systems, which could hit targets without human intervention or approval.
But, who is controlling the design, operation and target assignment to such ‘killer robots’? How such a robot will be able to understand the nuances regarding a complex situation and make life-threatening decisions? And many more.
The risk of bias and the need for transparency. AI systems learn by analyzing huge volumes of data and they keep adapting through continuous modelling of interaction data and user-feedback.
How can we ensure that the initial training of the AI algorithms is unbiased? What if a company introduces bias via the training data set (intentionally or not) in favor of particular classes of customers or users?
For instance, what if the algorithm responsible for identifying talented candidates from a pool of CVs, has inherited known or unknown biases, leading, for example, to diversity-related issues?
We must ensure that such systems are transparent regarding their decision-making processes. This is key to allow better handling of edge cases, while supporting the general understanding and acceptance by the wider audience and the society.
Access to data, knowledge, technology. In our interconnected world, a relatively small number of companies are collecting vast amounts of data. Access to this data would allow an accurate replay of our day-to-day life in terms of activities, interactions and explicitly stated or implicitly identified interests. Somebody with access to this data would ‘know’ our mobility history, our online search and social media activity, chats, emails and other online micro-behaviors and interactions.
An AI system will be able to ‘understand’ any online user — in terms of interests, daily habits and future needs; it could derive impressive estimations and predictions, ranging from purchasing interests to user’s emotional state.
If you think of this AI output at scale — analysing data at the population level — these predictions and insights could describe the synthesis, state and dynamics of an entire population.
This would obviously provide extreme power to those controlling such systems over this wealth of data. Just recall the Cambridge Analytica case. The data for a given individual user might be of low value, but when analysed at scale — for a sufficiently large group of users, with advanced analytical and inference models — it could drive massive socio-political influence.
The right to privacy. When you consider the possibility of unauthorized access to one’s online history (or other) data, the right to privacy is obviously at risk. But even in the case of an offline user — somebody who has deliberately decided to stay ‘disconnected’ — the right to privacy is still under threat.
Imagine a disconnected user (no smartphones or other devices aware of user’s location) moving through the ‘smart city’ of the future. A walk through a couple of major streets would be enough for the network of security cameras to capture user’s trails and possibly identify them via reliable facial recognition, against a centralized data store. There are obvious big questions on who has access to this information and under what conditions.
Unauthorized access and control. Security and access control is a critical aspect. If somebody compromises a smart system (for instance, an autonomous car) the consequences can be disastrous. Security of intelligent, connected systems and machines against unauthorized access is a top priority.
Technological unemployment. This is defined as the unemployment ‘explained’ by the application of new technologies — in the AI era it refers to the jobs replaced by intelligent automation.
In the years to come, we will witness significant changes in the workforce and the markets. Roles and jobs will become obsolete, industries will be radically transformed, employment models and relationships will be redefined.
For instance, tasks and activities related to customer care/call centres, document management, content moderation are increasingly based on technology and intelligent systems.
The same is true for roles related to operation and support of production lines and factories. Humans are being replaced by smart robots which can safely navigate the space, find and move objects (such as products, parts or tools) and perform complex assembling operations.
But AI proves to be very effective in handling even more complex activities — those requiring processing of multiple signals, data streams and accumulated knowledge in real time. A characteristic case is the autonomous vehicles which can capture and ‘understand’ their environment and its dynamics — they can ‘see’, decide and act in real-time. Professional drivers (taxi, trucks and more) will see the demand for their skill-set dropping rapidly.
Ethics, social responsibility and difficult decisions. AI enables optimal decisions in a real-time mode. Although in most of the cases the optimal decision is objectively determined and generally accepted, there are several examples raising ethical and moral issues.
For instance, an autonomous car which knows that it is about to hit a pedestrian, must decide if it will try to avoid the sensitive pedestrian via a risky (to its passengers) manoeuvre. And this needs to be decided in milliseconds.
The logic behind these critical decisions, must be predefined, well-understood and accepted. At the same time, the detailed history of activity and decisions of the autonomous car must be accessible and available for analysis — under certain data protection rules.
Disproportional power and control over data. Technology companies are investing heavily in artificial intelligence, both at the scientific/engineering and also at the commercial and product development level.
These corporations have an unmatched advantage when compared to any ambitious competitor out there. The massive datasets describing a wide range of human activity (searches, communication, content creation, social interaction and more), in many different formats (text, images, audio, video).
In an effort to retain their leading market positions, tech corporations tend to acquire those promising tech/AI startups disrupting the market. This could lead to super-powers, with a unique setup of AI technologies over massive amounts of accumulated user and machine data.
5. The promise
In the context of Internet of Things (IoT), billions of connected devices continuously send events, operation and other data, which are then processed by advanced Big Data, Machine Learning and Artificial Intelligence technologies.
This wealth of data, combined with the increasing ability to make sense of massive, complex data sets, is creating unprecedented opportunities for improvement across health, lifestyle, transportation, education and practically every human activity. Under certain assumptions, this technological revolution, will lead to a new era of prosperity, creativeness and well-being.
And yes, technological unemployment is a risk.
But in most of cases, Artificial Intelligence will play a supportive role to humans — empowering the human factor to perform better in handling complex and critical situations which require judgement and creative thinking.
In the future, humans will no more need to perform routine, limited-value, jobs. The workforce and the underlying employment models, will move from long-term, full-time employment agreements, to flexible, selective offering of services.
There will be a stream of new business opportunities empowering the culture of entrepreneurship, creativeness and innovation.
In parallel, numerous new roles and specializations will be created — focusing on technology and science, allowing people to free-up time from monotonous, low-value work, towards more creative activities.
Education systems will evolve to personalized programs and a life-learning mode. Innovation and creative thinking will be empowered by intelligent access of world’s accumulated knowledge, ideas and creative energy.
With the applications of AI in the Transportation industry, we will witness a significant reduction of accidents and fatalities on the roads. Moreover, people will benefit from lower transportation costs and increased level of service.
People will have better access to world’s digitized knowledge, with intelligent discovery tools. The ‘Fake news’ problem, along with content quality, security and safety online will all be improved by intelligent components and AI-powered services.
Artificial Intelligence is also improving our health systems: more accurate medical diagnoses, personalized medicine, shorter drug development cycles will significantly improve the overall effectiveness, level of service to patients and the general access to health services.
6. Getting ready
But how can we ensure proper use of Artificial Intelligence — in the interest of the individual and the society? How can we best adapt to the technological transformation which is already happening?
People need to achieve a general awareness and understanding of the technology, its potential, benefits and associated risks. Societies need to adapt to the new technology landscape and embrace Artificial Intelligence as a ‘smart tool’ helping people to achieve more. We all need to realize the value for humanity, but also see the treats from bad use of AI.
States need to adapt by modernizing laws, frameworks, social programmes and their education systems. New strategies are needed — to focus on education — along with new frameworks for the markets, businesses and social systems; they need to rethink how markets, companies and employment agreements should work in the new era of intelligent automation; they need to redesign the social mechanisms to cover a range of new scenarios and situations
People need to switch to a life-learning mode — learn to acquire new skills and explore new talents which are more relevant to the new order of things.
Thought leaders need to drive the right rules, frameworks and global agreements to mitigate the risk of centralization of power and control over data and technology.
This technological revolution brings great opportunities for prosperity and growth. We just need to somehow ensure that the technology will be applied and used in the right direction. We need a framework to guide the development of AI-powered applications with basic rules and those specifications that guarantee reliability, transparency and ethical alignment.
Key steps in the right direction are already happening. For example, there’s already discussion for banning ALWs. There’s also movement towards explainable AI (XAI) and the ‘right to explanation’. These allow understanding the models used for artificial intelligence (and how they make particular decisions — which is also required by the European Union GDPR — General Data Protection Regulation).
No comments:
Post a Comment