Robot in an Assembly Line. Photo Credit: Public Domain

There are many definitions of artificial intelligence in the same way as there are many definitions of intelligence. Intelligence is usually defined as capability to understand or capability to solve problems. Artificial intelligence would be the same capability built artificially for a computer.

The second definition is incomplete because any computer is mostly used to solve problems. Many times they do it much better than humans, but we do not think that they are intelligent if they are doing a pre programmed mechanical task. Intelligence is more linked to understanding than to solutions. In common life, we usually think that an engineer is more intelligent than an operator because the engineer has more knowledge and he understands the problem although the operator executes the task and provide the solution without understanding it.

There is no current computer system with the ability to understand a problem in the same way a human can do. However modern computer systems have the ability of adaptation to different circumstances related to the task to be executed. The process made by the computer to define a task to be executed is what defines the computer system as intelligent.

Intelligence is related to adaptability opposite to mechanism related to repeatability. A classic computer algorithm provides the same result for the same inputs. This is far from intelligence, and in the same way, any human system designed to provide repeatability is far from intelligence too.

Repeatability is a desired output in most industrial systems. In fact, industry is a result of the search for repeatability to provide a set of products of similar quality. The use of people in an assembly line is not an advantage instead of automatic machines to get repeatability. That is the reason why modern industries are more automatized.

From an industrial viewpoint, there is not a need to make a distinction between human and artificial systems, but between intelligent and mechanistic ones. Any industrial facility is a system composed by human beings and machines doing different tasks. Some of them are totally predefined and other ones require a new analysis and the adaptation of the processes. What makes a system intelligent is its capability to adapt the tasks to the problem instead of the degree of intelligence of its agents.

Artificial intelligence is not something fashionable. It is the logical evolution of any industry to the changing modern social environment. The environment of any organization changes much faster than a few decades ago. Organizations need to be more intelligent and less mechanistic. The age of quality standards has gone. Modern organizations live in the age of adaptability. Computer systems are not required to implement standardized procedures in an automatic way substituting human operators only. They are required to overcome the capability of decision makers to adapt the organizations the needs of the environment in a much faster way and in a much less complex one.

The limit of human intelligence to drive organizations is well known. The time of big data analysis and artificial intelligence support for decisions makers has come to make organizations more adaptable in modern changing environments. Artificial Intelligence exists to support intelligent organizations farther than human intelligence has got.

In any organization, AI will coexist with classic machines, with human operators working by procedures and with human decision makers. The design of the full system providing the required repeatability and adaptability will provide different organizations, with different degree of organizational intelligence. Although this cannot be measured, a satisfactory design can be evaluated through the measurement of complexity. A good resilience is a sign of proper adaptability and required repeatability for certain tasks.

Organizational intelligence is not an aim, it is a mean. The degree of organizational intelligence will depend on the industry: Nobody contracts an architect to hammer a nail. It should be very expensive and inefficient. And nobody contract a bricklayer to design a large tower. It should be very cheap but ineffective.

Advertisements

A simple algorithm to extract edges from the image Lena commonly used by computer vision scientists. Photo Credit: Public Domain

Large corporations are very interested in the new paradigm of Big Data. But what does it mean exactly? Basically, it is the acquisition and processing of large amount of data about markets in order to make reasoned decisions in a company. Big data can be used in order to improve any area of the company, from great strategic decisions to personalized offers for individual clients.

With this information we can see that it is a great opportunity for companies to improve their benefits. But, how must the use of big data be implemented?

The problem of big data is not the data. Of course, gathering data in not an easy task. A company must implement some source of acquisition in order to get the data. People were not prone to share their own information until social networks arrived. But, nowadays, we can get information about markets or clients from different sources.

The real problem to solve is how to process that great amount of data in order to get useful information for decision-making. The extraction of useful information from a large amount of raw data is a problem common with the technological area of robotics and computer vision. Mobile robots moving in a not structured environment must gather and process a large amount of data from vision sensors and so on, in order to determine a path or strategy in order to reach a destination point. A single image from a current common color camera is a signal 1024x768x3 bytes of information. This implies the processing of 2,359,296 bytes of information. A stereo vision system would use two cameras. If we work with video sequences at 25 images per second, we will need to process 117,964,800 bytes of information per second.

When we are working with big data, we need to diminish the size of data and to extract a more manageable subset of data especially useful for our interests. In computer vision, many times we work with edges. The mathematical meaning of edge is the derivative of the spatial sequence of data. In simpler words, we are interested in abrupt changes on the data that define different zones. This scheme can be used for any kind of big data problem. For instance, if we were analyzing geopolitical data in order to make decisions about the movement of our troops, we would be more interested in the changes of concentration of enemy troops around the borders instead of the internal situation of troops in order to prevent an invasion.

A single worker would last a lot of time in order to analyze the 2 Mbytes of information to extract the edge of an image; however, a computer can do it instantly. To program an algorithm to do it is not very difficult; however, it is not a task for someone who hated derivatives at the high school. Big data analysis is a matter of mathematical knowledge, not only a matter of writing code.

The knowledge required to analyze data fits well the knowledge of computer vision scientists among many other ones.

Chess Robot: Photo Credit Wikimedia Commons

The recent advances in Artificial Intelligence are provoking a debate about AI safety. There are some noticeable voices talking about the risk of the construction of intelligent robots. In my humble opinion as an expert in robotics from several decades ago, I consider that this problem is more philosophical than technical. The nature of knowledge and intelligence has been a philosophical problem from the beginning of philosophy, and there is not a philosopher that has proposed a satisfactory solution to it.

In the twentieth century, with the advances in automation, AI is born as a technical problem. The aim of AI is initially to endue automation systems with the ability to acquire some kind of knowledge about the controlled system or the environment in order to provide better responses than linear controllers. The initial successful AI systems are based on expert systems. Expert systems incorporate the knowledge of human experts and have some preprogrammed decision rules in order to improve the output of the controller.

The difference between a linear controller (a classical PID) and an expert system is that the former one is ruled by a mathematical algorithm which parameters can be precisely determined from the physical model of the system, and the latter one is ruled by a decision algorithm that managed the information deposited by experts into a database. However this kind of AI is fully predictable, the same inputs always provide the same outputs for the some stored knowledge into the database and that kind of expert systems cannot learn by themselves. The problem of expert systems is that they cannot cope with unexpected situations. They only can cope with situations expected by experts providing the knowledge.

When I was working in computer vision related activities as researcher, I realized that it was very important to have a philosophical vision of the intelligence in order to build more sophisticated systems, I began to read works by many philosophers:  Aristotle, Kant, Descartes, Hume, and so on. The problem of knowledge acquisition and intelligence is constant in the history of philosophy. Kant, considered the father of the modern philosophy, is a very interesting reading (and a hard one too) in this field. His separation of the concepts of phenomenon (object of sensible knowledge) and noumenon (object of pure reason) is very descriptive to be used in the engineering of intelligent systems. Kant is the first one philosopher that makes a discussion about the ability of the human brain to deduct two contrary logical solutions to several philosophical problems. He called them antinomies. In other words, there are some philosophical problems that cannot be solved by reasoning.

This is very important when we are talking about thinking machines. Natural intelligence is limited and most people talking about artificial intelligence are not conscious about it.

Current intelligent machines are much more complex than an expert system. They can acquire knowledge and learn. There are different techniques to get this with a computer, the most known technique can be the artificial neural networks. This kind of systems are bioinspired, they have a lot of little processing centers that can interconnect with their neighbors. Knowledge is stored in the weights of the connections between processing centers. The difference between ANNs and experts systems is clear. The output of the system depends not only on which is the knowledge to be acquired but on the how the knowledge is learnt.

There are recent examples about this fact, when some AI bots trying to learn from the interaction in social networks can finally transmitting unacceptable messages if a group of people interacting with the bot have provoked a learning driven to it. This is not very surprising because the same behavior can be easily provoked in a human child.

Why do I think that AI safety is more a philosophical problem than a technical one? Because we are trying to build a copy a natural system with a working model practically unknown and even that it is imperfect and limited.

The philosophical question here would be: Are we doing a copy of the human intelligence or are we doing a copy of the human stupidity? Actually, we do not know it, because we do not understand well what human intelligence is and how it is implemented, and what does a person more intelligent or more stupid.

However, the question about it we can make a safe humanlike AI system is more technical than philosophical, because safety measures can be included in the implementation of any computer system. This should be obvious for everyone. When we are not able to assure that our dog will attack people during our morning walk, we usually use a dog leash. The fact that a dog can move autonomously does not imply that we cannot limit its movements. In the case of robots, the leash can be incorporated internally by engineering design.

AI systems can be designed in many ways, with more or less degree of sophistication and it can include safety measures. AI will be useful in the industry if they are engineered to provide desired tasks and we avoid the raw copy of the human stupidity.

Christopher Columbus. Photo Credit: Public Domain

World has been always dominated by two different ideas about organization of groups of people: Either a strong leadership that drives the actions of different people or a leadership based on coordination where people provide opinions and the leader only extract the common wishes of people. In social governance the sides of these ideas are dictatorship and direct democracy. In the former one, only the opinion of a single person is taken into account. In the latter one, there is actually no leader that is substituted by a decision making methodology based on voting. North America is considered a place of democracy; however, a democratic organization in Columbus’ expedition never discovered it. Sailors, probably, would have voted to go back to Spain before reaching the coast of America.

Without considering moral affairs, both solutions could run a small society in a prosper way. Prosperity does not proceed from the decision making process but the decisions themselves. Methodology or leadership is not required in order to provide good decisions but in order to cope with increasing complexity. As societies grow both accepting the decisions of a leader and agreeing about many matters make themselves more difficult tasks. Men have developed mixed solutions where leadership is seen in different forms, and leadership and management methodologies must cohabit in an effective organization.

Governance evolves very slowly, however, business organizations evolves much faster. The number of formal management methodologies and theories about leadership is huge, although market finally selects some of them that are commonly accepted as good practices and they got to be extensively spread.

However, management consultancy is a business like another one. The success of its products depends not on their quality only but on the marketing activity too. Standards ‘de facto’ can exists due to a good marketing activity instead of the added value of the consultancy product, in a similar way as VHS video recorders overcame Betamax ones. Management consultancy can be many times driven by fashion instead of added value, and unfortunately what is fashionable is not usually driven by management professionals but communication experts.

In spite of these issues, the traditional two management visions are preserved although nowadays they have evolved in different management methodologies. A professional consultant usually knows several of these methodologies and he applies them depending on the preferences of the client organization. Clients demand consultancy products that many times are fashionable because they are not experts in methodologies and their general knowledge about it is based on the specialist press and management forums. Mass media drive fashions but we should not forget that they are private companies with their own interests and value chain.

The implementation of a certain management methodology must be considered for the long term, and it will not be changed in several years. That is the reason why managers should avoid fashions and center their decisions on the added value for their organizations. Most of them usually demand the implementation of ‘de facto’ standards to reduce the risk of their decisions.

An example of this discussion can be seen in software development. Traditional engineering projects are fully predefined and driven by an engineer that leads the project. This scheme fits well the construction of a building, a road or a bridge. Construction knowledge is in the hands of the engineer than makes decisions about the task to be made by the operators. When the complexity of the project increases the required knowledge implies the involvement of several engineers expert in different areas. The figure of project manager arises in order to lead the activity of engineering. In a large engineering project the knowledge can be very specialized making ineffective a scheme of decision making by voting, and a clear leader can help to make the process more effective. Direct democracy works well when the information about the problem is common and all people understand it equally, however, it does not fit well highly specialized groups because the success of the project may not be distributed equally among the knowledge of different experts. In this case, the project manager values how different tasks affect to the whole progress.

Software projects are different. In a software project, programmers have similar tasks related to programming, and the team can take advantage of sharing knowledge among them in order to improve their capabilities to do their own tasks. The scheme of full leadership is not efficient. Software engineering has developed different management methodologies, for instance, those ones known as agile. Agile methodologies preserve certain leadership for decision making; however, a lot of decisions are made by the team.

In Scrum, an agile framework that can be considered as a ‘de facto’ standard in software development management, the scrum masters are activators of the agile process. They are more facilitators and coaches than leaders in the traditional sense of this word. There is another role in this methodology related to management: the product owner. This role has some responsibilities related to important decisions about the progress of the project but is not a classic project manager too.

Different management methodologies exist because there are different kind of organizations with different aims and different characteristics. Successful organizations are those ones that choose a methodology that let them to cope with higher complexity, or in other words, those ones that reduces their working complexity in a more complex environment, making their processes more effective and more efficient.

Robot Dream Exhibition Hong Kong. Photo Credit: Public Domain

I cannot hide a little smile, when I can read at the press that “robots are coming to remove our job”. What a great discovery! Some “brilliant mind” has found that “evil” engineers are developing machines to avoid the hard work of people. Someone must think that the searching for a more comfortable life is not the sign of any civilization. Someone must think that the first civilizations do not invent livestock breeding in order to avoid the more dangerous, more laborious, less predictable, less effective and less efficient job of hunter. There is not any written record of some kind of group of hunters on strike. This is not a probable fact because unions were an invention in the recent centuries. However, there are archeological evidences that many tribes continued acting as hunter gatherers for a longer time until their disappearance. Not accepting the social technological changes is not only a matter of worse understanding of social evolution. It is a matter of driving the society to a subsequent disappearance in front of those ones that incorporate the technological advances.

I am not probably the most intelligent man in the world and I cannot see where the “evilness” of engineers can be found. Is the transformation of jobs into less dangerous activities for people evilness?  Is the transformation of jobs into less repetitive and more creativity activities evilness? Is a more effective and more efficient exploitation of natural resources evilness?

Christian-Jewish civilization is based on the premise that hard work is the result of the original sin of Adam and Eve (Genesis). For that civilization, hard work is a divine punishment. Then, it cannot be a social value. Perhaps, the “brilliant mind” behind the “evilness” of engineering is looking with religious eyes, because he sees it as a way to avoid a divine punishment. Suffering for a great cause can be considered an extremely great act of mercy; however, making people to suffer for something that can be easily avoided is only an act of masochism.

I do not want to be misunderstood. Effort and hard work related to achieve our own or even social aims must not be avoided. It should be promoted. What must be removed is the work that limits our capability to improve if there is a technological way to do it. Ancient Romans dedicated their lives to the defense and expansion of their empire while the repetitive work of cultivating their fields was done by slaves. War is a job harder than agriculture; however, the result of the war could not be put in the hands of slaves. There will be always many activities that will have to be done by ourselves, although those activities can be even the hardest ones.

I am conscious that people are not worried about the loss of jobs but about the loss of personal incomes, but this problem is not linked to engineering but linked to politics. It is a common social error in our days to translate the incompetence of political world to accept and to drive changes to the engineering world that is prepared to them from its beginnings. Engineering always provides solutions to problems, while politics creates them when politicians are not able to agree.

I can ask to myself if ancient Romans would have substituted their slaves for robots. That would be a great social problem. Before slavery all enemies were assassinated, in fact slavery was a social advance because it let to avoid the extermination of many people. The problem of robotics deployment can be not related to people jobs but to the change of the political order and the sharing of political power among rival political factions. Technology never is evil but people controlling it can be.

Robots are not the problem, are the solution. Robots are enemies of neither capital nor people. I would like that politicians against robots defending the “jobs of people” or a “stable economy” counted the number and quality of jobs before the use of computers and after it. Computers have provided more jobs, better labor conditions and higher benefits for companies. The arguments about the evilness of technological advances from any political side are neither reasonable nor defendable.