One of the major highlights in the world of artificial intelligence (AI) in 2018 was when it was used to generate realistic fake video footage by making the subject of one video mirror the motions and expressions of someone else in a different clip. That piece of discovery raised a lot of eyebrows, proving the notion to some extent, that in the wrong hands, AI could be a powerful tool for spreading misinformation. That theme also resonated at the recent UNESCO Forum on AI in Morocco where African countries gathered to chart a direction for the technology on the continent.
But, for Nigerian-born Victor Dibia, a human-computer interaction researcher and Carnegie Mellon University graduate, the benefits for Africa outweighs the bad.
“I think of AI (broadly defined as the science of imbuing machines with human-like capabilities) in ways similar to any effective tool or technology,” says Dibia in an interview with BusinessDay, “Like many influential technologies in recent times – telephony, mobile telephony, the internet, etc., the degree of their impact is determined by the degree to which they are effectively deployed.”
His latest work on creating African artworks using AI provides the basis for his confidence.
With the help of Google’s TensorFlow machine learning framework, Dibia trained a generative adversarial network (GAN) to generate images based on custom dataset – the African masks dataset. GAN is a two-part neural network (a computer system modeled on the human brain and nervous system) consisting of generators that produce samples and discriminators that attempt to distinguish between the generated samples and real-world samples. GANs have incredible potential, because they can learn to create worlds spookily similar to the real one in any domain including images, music or speech.
Data scientists at Alphabet recently tasked a GAN with generating convincing photos of burgers, dogs and butterflies.
“In August 2010, I had the opportunity to attend the 2018 Deep Learning Indaba where Google generously provided access to TPUs (v2) to all participants,” he wrote in a post.
His current work (Generating African Masks using AI Masks) explores the intersection of African art and AI. It addresses the basic question of how can machines learn aspects of human creativity and become tools that support creative endeavours. The work also doubles as an approach to draw attention to the rather underrepresented but deeply rich area of Africa art.
“Early results suggest that an AI model is able to learn concepts around the design space for African masks (textures, geometry) and can generate new interpretations of what these artistic pieces could be. It contributes to the emerging area of computation art, or generative art by providing a computational lens to evaluate the styles of these art forms and enable conversations around them. In recent times, we have witnessed legitimate commercial interest in this area, with an AI – generated art piece sold at a reputable auction house such as Christie’s. It will be quite refreshing to see this level of engagement with African art.”
In October, 2018, a portrait produced by artificial intelligence “Edmond de Belamy, from La Famille de Belamy” sold for $432,500 including fees, over 40 times Christie’s initial estimate of $7,000-$10,000. The New York Times reports that the bidding lasted just under seven minutes, during which the buyer competed against an online bidder in France, two other phone bidders and one person in the room in New York. When the hammer came down, the bids had reached $350,000, the fine price before fees.
GANs have been used in art since 2015 by artists such as Mario Klingemann, Anna Ridler, and Robbie Barrat.
Dibia says his work ensures that African art finds a voice in the growing research area of generative art and ensures it benefits from new conversations.
“The most impactful and popular form of artificial intelligence today follows a paradigm known as “supervised learning” where a machine is able to independently learn without any explicit programming, simply by looking through massive amounts of labeled data. For example, a machine is able to identify objects in an image (e.g. tables, chairs, cars etc) simply by looking through a dataset of images that have been labeled as containing each of these items. This process is known as training a model,” He says.
He acknowledges that the fear of job loss as a result of AI and automation is real. Africa has a high risk exposure with up to 85 per cent job loss predicted in parts of the continent. But with the right strategies and programs, Dibia says Africa which has one of the youngest populations in the world can weather the storm ahead.
However, government and private sector will need to collaborate to yield the most compelling results.
“Such a partnership will allow each entity leverage their respective strengths and achieve results that neither individually can,” says Dibia, “The private sector, unencumbered by bureaucracies, can provide leadership in research, innovation, and rapid execution of ideas. The government can provide long term policies, infrastructure and governance that support AI research. Some examples of these programs include work being done by Co-Creation Hub in Nigeria where they have partnered with government institutions on multiple health care and social good projects.”
It is important to note that opportunities also exist for AI to enable aspects of healthcare – examples include low cost disease diagnostics, reducing infant mortality, applications in telemedicine – agriculture (crop yield predictions, crop disease predictions), personalised education, citizen participation and transparent governance.