Intro to AI Bias

Sophia Will Not Destroy You. But...Dr. Frankenstein?

It's been five years since Sophia the Robot first debuted in mainstream news outlets as the robot who wants to destroy all humans. It was 2016. Hanson Robotics had just introduced its first human-like android, Sophia. In an iconic interview between creator and creation, David Hanson asked the android if it wants to destroy all humans to which it answered, "Okay, I will destroy all humans." Forgetting any semblance of basic English grammar and looking past the incredibly awkward, mechanical facial movements of the robot, the world went into a sensational panic. The age of machines taking over the world was finally here!

Except that it wasn't. It isn't. But the question still remains at large: Will we, one day, be taken over by robots? To that I answer "No." The day when robots have a form of consciousness will be the day that God comes down and puts it into our machines. Now, I know that there is much controversy over this, but with the technology out there now, I stand on the side that we are so very far away from artificial intelligence coming remotely close to becoming organic entities of their own. (Who knows, though, maybe there is already some way advanced technology out there, hidden in Area 51.... ;D I'm just kidding. The whole point of this post is to relieve the panic about the science fiction fantasies floating in our brains, over-hyped by mass media hysteria.)

However, I do think that regardless of how close these technologies come to being organic the only real fear lies in how they are built to interact with the world in the first place. Even if they became more intelligent than humans, the scientists behind the technologies have full control over what our futures will look like... Before you go into a panicked state about an impending apocalypse, though, let's take a step back. For those who are not familiar with the tech world, I've broken down this topic into three concepts: artificial intelligence, machine learning, and robotics.

Artificial Intelligence (AI)

Artificial Intelligence, or AI for short, is the study of how well computers, machines, or in other words, inorganic entities, can mirror intelligence, which is often associated with its likeness to human thinking. Thinking includes all forms of human cognition, from the logical to the emotional to the irrational and beyond.

Within the field of AI, the question of intelligence revolves around the caveat that intelligence itself is extremely difficult to define. It is a term that has been grappled with not only recently within the world of AI, but also historically within psychology, sociology, education, and so on and so forth. So, if you've got a difficult time pinpointing what AI really is, you might actually be more onto something than those who've claimed they've nailed it on the head!

One of my favorite imageries from an artificial intelligence course I took is that the brain is a magical, black box. Scientists have been hacking away at this box, only to find more depth and questions that further reveal a black hole. An example of our brain's mysteries is its incredibly quick ability to recognize patterns. Pattern recognition is just one of the many capabilities that machines are not close to mimicking. This should give you an idea of how difficult it really is for AI to reflect the human mind and all of its magic, considering we barely have a hold on the human mind itself.

Chiwawa or Muffin?

Machine Learning (ML)

Machine Learning, or ML for short, is a subfield of Artificial Intelligence, and can be considered as one of the tools for implementing said AI. More specifically, it is a data-driven approach for computer learning, the act of achieving some form of the aforementioned intelligence. Now don't be fooled by the term learning. Scientists personify the machine and software as a method for talking about the product and its abilities, but we have to be cognizant of when it is literal and when it is figurative -- and, so far, it's usually been figurative.

Most ML depend on data which is cleaned and pre-processed, or turned into a digestible form, by an algorithm, or a set of rules. This data is then fed into and interpreted by a model, which is often a statistical or mathematical function that produces an output based on the input data. A good model will have been trained by, or exposed to, a dataset that prepares the model for the recognition of unfamiliar inputs. The technicalities of each of the above terms can be further explored in later discussions, but can be understood quite literally for what they sound like for the introductory purposes of these concepts.

The basic idea is that a machine learning algorithm learns from the data. But, remember, that this learning, is far from what we imagine the human brain does, though there are forms of ML that try to reflect, as closely as possible, things such as how neurons in our brains fire to properly recognize a photo of Beyonce, let's say.

With that said, the technologies and approaches in ML are changing constantly and can vary within the realms of natural language processing (NLP), which is the study of computers' abilities to process natural human languages; computer vision, which is the study of how computers learn from images and videos; speech recognition, which is the study of how computers detect auditory input and decipher the language being spoken to it; and more. The ML portion of all of these subfields of study is the algorithm that is built to take in an input (i.e. a set of texts, a set of images, or a set of voice recordings) and create an output that aligns with the engineer's goals.

Robotics

Robotics is a field of its own that combines software and mechanical engineering to create physical entities that can partake in certain decisions and actions. These machines can be programmed (the software portion) to do (the mechanical portion) tasks such as pouring orange juice into a cup every morning at 8 AM.

However, the field of robotics has grown beyond simple, redundant tasks. It is incorporating artificial intelligence to create more responsive machines that can not only perform limited and repetitive in-built features such as the robot that is programmed to pour orange juice into a cup every morning at 8AM, but now can also interact with and react to the environment around them. Take this orange-juice-pouring-robot, for example. If we wanted, we could now re-program it with AI to recognize patterns such as "nobody drinks the orange juice at 8AM on rainy days" and, consequently, the robot will begin to stop pouring orange juice on rainy days.

The door-opening-robot is another great example of robotics. If not powered by AI, it might be a simple entity that can do one task -- open doors and walk through them. More specifically, it might be programmed to only be able to handle specific types of doors: ones where the handle needs to be turned and pulled rather than having to be pushed or even unlocked. However, the above pictured robot most likely is powered by some form of AI which allows it to walk around, and understand what to do when it encounters any door.

Presumably, through the combination of computer vision, which as mentioned before is the study of how computers learn from images and videos, and through effective sensory-motor hardware, this door-opening-robot is able to maneuver situations that involve doors. The robot's ability to perform such a seemingly simple task is dependent on meticulous calculations as well as the existence of a variety of environmental scenarios all of which are part of the ML software that controls the robot's moves. The more complex the environment (say, we add stairs, or a puddle), the more data it needs to process, and be engineered to handle, physically.

. .

I hope that this review helped a little, just to get the ball moving in terms of conversing further about this topic. For those who are already familiar with the technology, thanks for baring with me (and if you have more to add to what I've laid out above then I welcome your feedback!). I want the conversation on AI and robotics to be open to everyone, regardless of their level of expertise because the technology affects and will continue to affect all of us. The last thing I want is for people to be deterred from being a part of the conversation due to their lack of a technical background because in our increasingly technological world, we should all have a say in what is getting built around us.

Taking a look at these technologies, and where we are now, even with advanced machines such as Sophia the Robot, Tesla's self-driving cars, or police digidogs, I don't see a future in which these entities conscientiously take over the world. The way that we, humans, process information in our tiny brains is infinitely more efficient yet complex than anything out there today. With that being said, yes, I do foresee a grim future, overcome by robots commanded by the wrong people. In this grim future, the faces behind these machines aren't necessarily evil. They're just human and neglectful. But we don't really live in a world where we can afford to be neglectful. And, we see the harm being done already. Sophia might not be here to destroy you. But there are Dr. Frankensteins out there who need to take responsibility for the social and cultural implications of their creations.

The conversation about technology cannot be held without considering ethics. Thankfully, the debate on AI bias has been of greater interest to politicians and thought leaders. For those who are unfamiliar, AI bias refers to the innate human biases that are built into artificial intelligence machines by the engineers. These biases might not even be intentional, but the impact? tremendous. An example of where AI bias has been a hot topic is in how it is used in recruiting: in 2018, Amazon was under serious scrutiny when it was revealed that it had been using a hiring tool that favored male candidates over female candidates. Another controversial use of AI came to light when cities across the US started using machine learning algorithms to predict where crime is more likely to happen and who is more likely to commit them. Just as with the recruiting algorithm, the supposed crime fighting algorithms were likewise biased.

Josie Young warns inventors and users of the issues that AI pose for the female community including but not limited to the normalization of abusive language toward female assistants

The consequences of these algorithms is not only in how they directly affect individuals, but also in how they directly affect entire generations. Ever wonder why Siri, Google, and Alexa all have female voices? I wonder that too. Is it because makers believe it is more pleasant to hear a woman's voice? If so, to the ears of whom? All over the world, today, there are people barking commands, left and right, to their female voice assistants. Now, imagine how that affects our psychologies, and how that would influence our behavior toward actual people. In particular, this is a very gendered issue that Josie Young addresses in her TED Talk, Why We Need to Design Feminist AI. In her talk, she warns inventors and users of the issues that AI pose for the female community including but not limited to the normalization of abusive language toward female assistants. And, she strongly suggests that AI should be fixing society's problems, not adding to them.

The problem is not only what is being created but also who is reacting to and interacting with these new technologies. The first headline that appears in my search for news about Sophia the Robot reads, "Hot Robot at SXSW Says She Wants to Destroy Humans" (referring to the first debut interview between Sophia and Hanson.) This is one of many examples of how even androids are objectified and subject to the same gendered expectations of humans, and it goes to show that sexism is still out there, embedded into our culture and our language. Sophia the Robot is even asked questions about motherhood, which sparked yet another sensational outburst: "Robot Wants Babies!" The robot isn't even capable of reproduction nor would another robot need "parenting", but we find that the world is so fixated on personifying these androids, and, furthermore, cannot detach societal ideals of gender from them that these irrelevant yet harmful conversations keep coming up.

The EEOC is keenly aware that these tools may mask and perpetuate bias or create new discriminatory barriers to jobs

Thankfully, governments have begun to take some action regarding the ethics of artificial intelligence. In late October this year, the U.S. Equal Employment Opportunity Commission announced its "initiative to ensure that artificial intelligence (AI) and other emerging tools used in hiring and other employment decisions comply with federal civil rights laws that the agency enforces." EEOC Chair Charlotte A. Burrows says, “Artificial intelligence and algorithmic decision-making tools have great potential to improve our lives, including in the area of employment. At the same time, the EEOC is keenly aware that these tools may mask and perpetuate bias or create new discriminatory barriers to jobs. We must work to ensure that these new technologies do not become a high-tech pathway to discrimination.”

The further development of AI is an all-hands-on-deck project. Engineers, consumers, law makers, educators, investors alike need to be intentional and proactive! While we don't need to concern ourselves with being ruled by technology, we do need to free the world from biases, stereotypes, and backwards thinking because if we don't, we'll only be building a worse world for future generations to come.

till next time,

michelle