Can an A.I. system be a designer? (Part 2)
A deeper understanding of AI & ML in context of design systems
Previous: Can an A.I. system be a designer?
Introduction
Last week I wrote an article that examined the intersection between the capabilities of Machine Learning (ML), Artificial Intelligence (AI) and designers. The purpose of this article is to understand why and how popular machine-learning algorithms are being used- as well as some ethical considerations for designing with ML. This information should help designers communicate with those specialized with ML and AI, especially as they become more and more relevant in consumer technology.
ML Algorithms — The power behind AI
Below is a list of some different algorithms as well as the systems that they are selected to emulate. Data in this section are drawn from Ayman Youseff’s 2016 article The role of artificial intelligence in photo-voltaic systems design and control: A review. I’ve also included links to videos explaining how these algorithms work and why they are used.
- Neural Networks (NN) are algorithms that mimic the patterns of the human brain. The human brain is the most powerful and efficient calculating machine with learning capabilities. It can solve real problems that the ordinary PC can’t solve. (video)
- Fuzzy Logic (FL) mimics the method of human thinking. It depends on linguistic variables with IF-THEN rules base. (video)
- Simulated Annealing (SA) is an evolutionary algorithm that emulates the
nature process of gradual cooling. The algorithm is used for optimization problems. (video) - Genetic Algorithm (GA) mimic the natural behavior of evolution. In this
algorithm there is an initial population of genes. A fitness function is calculated to find the best genes in the population. The genes with the best fitness function values are chosen to produce the next generation of genes. (video) - Ant Colony (ACO) are algorithms inspired by the behavior of ants for finding the shortest path for collecting the food to their nests. Ants move randomly to search for the optimum path. They lay down pheromones as they move. Finally the ants follow the path of higher density of pheromones. (video)
- Particle Swarm Optimization (PSO) This is an optimization algorithm inspired by the swarm and flocking of birds. It has a wide range of applications due to its simplicity, high performance, and flexibility. It simulates the behavior of a group of birds searching for food. Each bird moves in a direction based on its experience and group movements. (video)
- Adaptive-Neuron Fuzzy Inference Systems (ANFIS) This is a hybrid technique that combines the benefits of both neural networks and fuzzy logic. (video)
I once had the opportunity to work with a zoologist at Balance who was designing a tool that was inspired by the movement patterns of the whale. We humans build machines to do things that we see being done in the world. by animals and people, but we typically don’t build them the same way that nature built us. (Brynjolfsson & McAfee) Many of these algorithms above are examples of biomimicry — the design and production of materials, structures, and systems that are modeled on biological entities and processes.
Radiology — A science wondering the same question
Radiologists face a similar anxiety of technological unemployment since AI algorithms are helping to detect breast cancer (Dheeba, et al.), colonic polyps (Summers, et al.), and pulmonary nodules (Chen, et al.), with many more applications to come. Tomer Nawrocki wrote an article called Artificial Intelligence and Radiology: Have Rumors of the Radiologist’s Demise Been Greatly Exaggerated? (2018) where he explained how although AI is unlikely to completely replace the radiologist, a new breed of software applications based on machine learning is poised to relieve radiologists of many tedious, repetitive, and time-consuming simple tasks, which will lead to increased productivity.
Comparing radiology to design is beyond the purpose of this article, but we can see an overlap in how A.I. can be used to support the designer or radiologist.
Of course this all sounds nice, A.I. helping humans. But how do we deal with the unpredictable nature of this technology? Chaotic data glitches? Bias? False-positives/false-negatives? and, a little more far-out- Singularity? A lot of Science-Fiction serves as a harbinger for the abuse of technology. George Orwell, William Gibson, and others have described dystopian scenarios involving the loss of freedom and the use of technology to empower despotic rulers and control information flows. (Brynjolfsson & McAfee)
Human-Centered Machine Learning — The Designer’s Answer
In July of 2017, Google designer Jess Holbrook released the article Human-Centered Machine Leaning on Medium with seven tips to stay focused on the user when designing with A.I. It’s part of a philosophy in the Google UX community called “human-centered machine learning,” where machine learning algorithms solve problems while keeping human needs and behaviors in mind.
- Don’t expect Machine Learning to figure out what problems to solve. Many companies and product teams are jumping right into product strategies that start with ML as a solution and skip over focusing on a meaningful problem to solve.
- Ask yourself if Machine Learning will address the problem in a unique way. There are plenty of legitimate problems that don’t require ML solutions.
- Fake it with personal examples and wizards (for testing). Use personal examples from participants and Wizard of Oz studies for testing the ML solutions.
- Weigh the costs of false-positives and false-negatives. Your ML system will make mistakes. It’s important to understand what these errors look like and how they might affect the user’s experience of the product.
- Plan for co-learning and adaptation. You want to guide users with clear mental models that encourage them to give feedback that is mutually beneficial to them and the model.
- Teach your algorithm using the right labels. As UXers, we’ve grown accustomed to wireframes, mockups, prototypes, and redlines being our hallmark deliverables. Well, curveball: when it comes to ML-augmented UX, there’s only so much we can specify. That’s where “labels” come in.
- Extend your UX family, ML is a creative process. Think about the worst micro-management “feedback” you’ve ever received as a UXer. Can you picture the person leaning over your shoulder and nit-picking your every move? OK, now keep that image in your mind… and make absolutely certain that you don’t come across like that to your engineers.
Though Google’s guidelines are not universal, they assist in answering a few simple but important question that everyone involved in the future of design should be asking:
How do we make sure A.I. benefits humanity? How will we design machines to understand the same societal lessons that humanity has learned? or, as Katharine Schwab mentioned in her article Google’s Rules For Designers Working With AI, how do we design algorithms that aren’t evil?
That is a topic for another article.
Conclusion
I hope this provided a broader look at Artificial Intelligence and it’s modern applications. It’s good in design to know what different A.I. systems are and why they are used. The technology is still far too inaccessible and infant to understand the potential for how they are used in design. This is why it’s really important to rely on a system of ethics and leadership in the design world to provide guidance when approaching this new and powerful technology.
Next: Can an AI system be a designer (Part 3)
Sources
Brynjolfsson, E., & McAfee, A. (2016). The second machine age: work, progress, and prosperity in a time of brilliant technologies. New York, NY: W.W. Norton & Company, Inc.
Chen, Sheng, et al. “Development and Evaluation of a Computer-Aided Diagnostic Scheme for Lung Nodule Detection in Chest Radiographs by Means of Two-Stage Nodule Enhancement with Support Vector Classification.” Medical Physics, vol. 38, no. 4, 2011, pp. 1844–1858., doi:10.1118/1.3561504.
Dheeba, J., et al. “Computer-Aided Detection of Breast Cancer on Mammograms: A Swarm Intelligence Optimized Wavelet Neural Network Approach.” Journal of Biomedical Informatics, vol. 49, 2014, pp. 45–52., doi:10.1016/j.jbi.2014.01.010.
Holbrook, J. (2017, July 09). Human-Centered Machine Learning — Google Design — Medium. Retrieved February 25, 2018, from https://medium.com/google-design/human-centered-machine-learning-a770d10562cd
Nawrocki, T., Maldjian, P. D., Slasky, S. E., & Contractor, S. G. (2018). Artificial Intelligence and Radiology: Have Rumors of the Radiologists Demise Been Greatly Exaggerated? Academic Radiology. doi:10.1016/j.acra.2017.12.027
Schwab, K. (2017, September 13). Google’s Rules For Designers Working With AI. Retrieved February 25, 2018, from https://www.fastcodesign.com/90132700/googles-rules-for-designing-ai-that-isnt-evil
Summers, Ronald M., et al. “Automated Polyp Detector for CT Colonography: Feasibility Study.” Radiology, vol. 216, no. 1, 2000, pp. 284–290., doi:10.1148/radiology.216.1.r00jl43284.
Youssef, A., El-Telbany, M., & Zekry, A. (2017). The role of artificial intelligence in photo-voltaic systems design and control: A review. Renewable and Sustainable Energy Reviews, 78, 72–79. doi:10.1016/j.rser.2017.04.046
