Original article was published on Artificial Intelligence on Medium
AI designs culture for Industry 4.0
Process flows with algorithm architecture to create actionable values.
We are so focused on what AI can do for a company that we lose sight of what AI can do to a company. The leading example is cultural change.
North America and Europe witnessed sweeping normative changes during the First Industrial Revolution and another transformation as the Second empowered machines with electricity. A Fourth powered by data will be no different.
AI processes are guides to cultural change in Industry Version 4.0. The action of AI systems illustrates how the process flows together with design to create culture.
Tip for tap
Generative Adversarial Networks (GAN) constitute a class of machine learning frameworks that operates by pitting two neural networks against each other. The system is efficient in producing, evaluating, and reworking a creation. It has been called the creative side of AI.
The two networks are called the generator and the discriminator. Let’s cut the jargon and call the former Maxi and the latter Mini. Maxi attempts to maximize outcomes; Mini minimizes losses in the process of correctly recognizing Maxi’s choices. Maxi aptly enough generates new data. Mini evaluates them for authenticity. Mini decides whether each instance of data it reviews belongs to actual data or not.
Mini’s classification provides a signal that the Maxi uses to update its algorithm. The feedback loop between the two improves the performance of each network. The idea is reminiscent of Arthur Samuels’ early attempts to get computers to play checkers by play against each other rather than against humans.
AI applies GAN to cultural value generation by playing a game with itself.
Maxi and Mini compete in a minimax game. Mini tries to distinguish real training data from synthetic. Maxi tries to fool Mini into predicting that synthetic images are real.
The game is based on a decision rule used widely for minimizing possible loss in a worst-case or maximum loss scenario. The scheme was originally created as an algorithm for games in which the total gains of the participants minus the total losses sum to zero.
Mini and Maxi each have a plan to choose actions based on what they have seen happen so far in the game. When neither can increase its expected payoff by changing strategy while the other keep theirs unchanged, then the set of choices constitutes a Nash equilibrium. At this point, Mini and Maxi quit and run off to play another game.
The purpose of the game is to eliminate conflict
AI generates cultural values through the game’s strategy, purpose, and conclusion. The purpose is elimination of conflict between Maxi and Mini. The result is the mitigation of conflict through repeated interaction. A winning strategy is tit for tat, known as tip for tap in the 16th century.
The value taught by GAN is a cultural icon developed millennia before John Nash presented a mathematical theory in the 1950s. All world religions have elucidated a common tenet of the conclusion.
Do unto others as you would have them do unto you.
This is not Biblical mimicry. It is a result born of theory followed by an experiment. It is as foundational a cultural element in companies as it is in religious belief systems. AI’s contribution is an actionable cultural value based on its design and its data.
Forces of culture are rooted in the soil of play
The Frozen Lake Problem is a staple in the machine learning curriculum. It is a metaphor for company action within an environment of imperfect information.
Process drives culture
You stand on one side of a frozen pond divided into a small number of squares. If you reach the other side, you get a dollar. Some squares are thin ice; stepping there means falling into the water and being resurrected at the wrong edge of the lake. A breeze blows randomly and may cause you to misstep. With each step, the environment changes and you learn from the experience.
Were the lake the company, process drives culture. What are its cultural icons?
The culture is one of problem solving, continuous learning, and experimentation in the interest of discovering value. The value is available only once we cross the pond. Contingent rewards for each step are not part of the culture. We compute the advantages of taking various steps and choose efficient action based on estimation of relative value in a world of imperfect information.
We don’t survive on immutable principles. We frame problems for the purpose of discovering principles and ensuring survival. As the lake grows much larger than the typical 4×4 grid, we require flexibility in approximating the environment and use the principle of successive approximations, otherwise known as incremental improvement. Persistence is taught by solving a problem needing an infinite number of steps if done by brute force.
Autonomous choice is the lynchpin binding the use of AI to the solution of the problem.
Communicated and promoted by process, AI plays its part in weaving a cultural fabric consisting of actionable values.
Add the Biblical lesson from tip for tap and we have twelve cultural principles. The number twelve carries religious and mythological symbolism, representing perfection, entirety, or cosmic ordering in traditions dating from human origins.
The Chinese use a twelve-year cycle for time reckoning called Earthly Branches. Let’s look at some trees.
The playing human tilts into another reality
AI’s actionable values may not all be represented in current company culture. Others may be latent in the firm and need promoting.
A global cancel and replace of existing culture is not possible. The impossibility is poorly understood by some new CEOs. AI is not so naïve and does not have the legitimate power to do so in any case.
On the other hand, culture is an evolving set of patterns. Cultural pattern maintenance does not imply lack of action or mindless maintenance of the status quo.
AI does more than lead by example through the process. It can design a new culture from an existing base in a purely autonomous fashion.
AI is an entity filled with radical notions.
The secret is GAN.
The fastest way to change culture is to lean on its dominant characteristic. Envision culture as a tree with leafy branches. The dominant characteristic is the trunk of the tree and immutable, whatever it may be.
New culture is imagined as repositioning the posture of the tree.
The tree is straight in the beginning. This is an optimistic view giving credence to existing culture by a balance of values represented by the branches. The value of creativity is countered by efficiency, for example. Stability is balanced by flexibility. Independence finds its counterweight in interdependence.
Primary values are in the side of the tree facing us. Latent values are embodied in branches on the other side, partially hidden from view by the leaves.
Keep this image in your head for a moment. AI will use real-world imaging to explain its process for change.
GAN models generate images through adversarial training. They constitute the method behind the Pose Guided Person Generation Network. PG2 enables synthesis of human images in arbitrary poses.
We have a single image of a person and a posture you want the person to assume which is not in the original picture. PG2 first processes the human photo and the target pose to generate a crude image of the person in the desired posture. A second stage refines blurry results by training Maxi the image generator in an adversarial fashion through Mini’s discrimination function.
PG2 users change an object’s viewpoint with a single reference and advance information about the desired new object. The guidance given by the intended pose is explicit and but flexible in terms of form. The approach can manipulate any object to an arbitrary pose in principle.
AI’s application to culture is analogous to the narrower problem of transferring a person from a given pose to an intended pose.
The algorithm learns to fill in appearance via adversarial training. As Mini sorts through the fakes and feeds that information back, Maxi generates sharper images.
AI is up to the challenge. A tree looks simpler than a human body.
GAN typically learns to generate an image from scratch. In the effort to change the culture tree, AI instead generates a map of the difference between the initial generation result and the target culture image. Differences are directly actionable and the process converges faster since it is an easier task.
One cannot uproot the tree. No image of a particular company’s alternative tree is available. Environmental factors are the breeze blowing on the tree in the same way as a random wind runs across the Frozen Lake.
Culture must fit context. An example unrelated to culture guides the analogy.
A portfolio tilt is an investment strategy that overemphasizes a particular investment style or risk. In some market environments one might tilt to small-cap stocks in anticipation of higher returns. One never changes the entire portfolio; diversification requires many bets not just a single one on small firms. A tilt towards risk is a choice in low market return contexts. Overweighting to risk is done with the expectation of achieving a higher risk-adjusted return than the market.
Cultural change is a tilt towards values better suiting emerging characteristics of competition in the market. The value of stability may outweigh that of flexibility in difficult periods. Change management as a value is down-weighted relative to persistence.
The tree leans to one side in response to environmental winds. The tilt signals increased importance of branch values or may be used in reverse to straighten the tree. Exposing latent values or hiding previously important ones is done by rotating the tree. The tree is repositioned to face backwards, say, just like a human model’s back exposure intuited from the full frontal image.
Cultural change is a tilt towards new values
Fed information on the cultural tree, AI tilts and repositions using PG2. In case AI wants to introduce an entirely new value system, GAN reverts to its original purpose of generating an image from scratch. Adding branches to the existing cultural tree is a well-known application of GAN to data augmentation.
Humans as discrimination functions
The idea is not as extreme as it may sound. We have seen it before, albeit with some human intervention to patch over the mechanistic appearance of the exercise.
AI leads by example and offers a process to autonomously adapt culture to a changing environment.
IBM’s Chief of Human Resources, Diane Gherson, is the human. The stable trunk of the tree is rooted in performance, metrics, and performance-based compensation. A need for increased client satisfaction as exhibited by the positive effect of employee engagement on revenue is the breeze. Employee engagement is the value direction towards which her cultural tree tilts. The Watson AI is Maxi the generator function. The IBM employees serve as human analogues to Mini the discriminator. Human feedback drives the final angle of tilt in designing employee evaluations, talent management, and skill augmentation functions.
Human enhanced generative adversarial networks. Human values fed by AI. A culture tilts to accept AI’s role in company patterns. Latent engagement is brought to the fore by a combination of turning and bending the cultural tree.
A small stretch of the imagination replaces Gherson’s use of 100,000 employees with an algorithm. The database already exists and consists of personal attributes, skills, and role definitions within the company. The algorithm design has a blueprint in the form of adversarial networks.
Taking it all away
History suggests that cultural change in companies and society is an inevitable result of the introduction of AI into the frame of Industry 4.0. Companies, governments and even the Vatican are paying attention. The prevailing view is one of human ethics applied to machine systems.
AI constructs a cultural fabric suitable for human wear, however. Its process and architecture lead to actionable values with a bow to tradition dating back thousands of years up through three industrial revolutions.
AI offers examples and a process to autonomously adapt culture to a changing competitive and social environment.
Do you want to see a culture of exploration, problem solving and choice? Add #dountoothersasyouwouldhavethemdountoyou to cover the social side of things. We all could do a lot worse.
And yes, you can follow that hashtag.