AI Lexicon — B

Big data

Big data is data with larger, more complex data sets containing huge amounts of information. AI algorithms which are designed to create predictions (predictive AI) are only as good as the data they are trained on — big data has allowed algorithms to more accurately generate and predict outcomes.

In their book, Big Data: A revolution that will transform the way we live, work and think (2013), Viktor Mayer-Schönberger and Kenneth Cukier write that big data was "an important step in humankind's quest to quantify and understand the world […] Big data is all about seeing and understanding the relations within and among pieces of information that, until very recently, we struggled to fully grasp."

These are two key points. Many other experts agree that it is not the size or amount of data but the complexity and multilayered nature of the data — what some people refer to as "messy" data — that makes the difference. In addition, it's the tools we use to analyze the data and the interpretations we draw from them that count. (za/fs)

Bias in AI

There's been bias in artificial intelligence systems for as long as those systems have existed — after all, AI systems are designed and built by humans, and we all have biases, some of which are unconscious.

So, when, for instance, police get programmers to build a facial recognition system to help them spot potential car jackers or drug dealers, they will look at common characteristics for car jackers and drug dealers — people they have arrested and have been convicted for a crime in the past. Those characteristics may include things such as their wearing hoodies, having tattoos, or the color of their skin.

General awareness of bias in AI grew in 2018 when two US-based researchers found "gender and skin-type bias in commercial artificial-intelligence systems."

They detected an error rate of 0.8 percent for light-skinned men and 34.7 percent for dark-skinned women in three general-purpose facial-analysis systems. In plain language, it seemed that AI facial recognition systems were unable to detect distinguishing features or contours in the faces of people of color — especially, the darker the skin.

The paper was published by Joy Buolamwini, then a research assistant at MIT Media Lab and self-styled "Poet of Code", and Timnit Gebru, who at one point worked an ethics researcher at Google, where she looked at the risks in large language models, until she was reportedly "forced out."

Their findings were considered especially significant because they suggested that gender and racial biases were "baked" into the AI systems in their research.

Such biases can grow quickly if the majority of research is done by white men — this is not just a cliché of reverse discrimination, but still often the truth — using white Western ideologies and samples of people and things.

AI bias arises in AI tools when they are trained on datasets which do not contain diverse information. In the case of facial recognition algorithms, early versions were trained on people predominantly from European backgrounds, so never learned to accurately recognize people from other backgrounds. (za/fs)

Brain-Computer Interface (BCI)

A BCI is device which connects the brain to a computer. It consists of a “chip”, a small array of electrodes, which is surgically implanted into the brain. Some of the electrodes stimulate the brain with tiny amounts of electrical current to change neural activity and elicit sensations. Other electrodes record from the brain’s activity.

The chip’s input to the brain and output from the brain is controlled by a remote computer. AI has been instrumental in interpreting the complex patterns of brain activity and controlling brain chips.

Examples here include a BCI by Neuralink, a company partially owned by Elon Musk, which is being tested in humans to help people with neurological conditions. Neuralink is by no means the only company using BCIs. BCIs have been tested in humans for several decades and have allowed people with paralysis to feel and move their limbs, and help people communicate again. (fs/za)

 

Sources:

Big Data: A revolution that will transform how we live, work and think by Mayer-Schönberger, Viktor; Cukier, Kenneth (John Murray Publishers, 2013)

What is big data? (Global Investigative Journalism Network / Jenna Dutcher, September 9, 2014) https://gijn.org/what-is-big-data/ (accessed September 18, 2023)

Joy Buolamwini (profile page, MIT Media Lab) https://www.media.mit.edu/people/joyab/overview/ (accessed May 14, 2024)

Study finds gender and skin-type bias in commercial artificial-intelligence systems (MIT Media Lab) https://news.mit.edu/2018/study-finds-gender-skin-type-bias-artificial-intelligence-systems-0212 (accessed May 14, 2024)

We read the paper that forced Timnit Gebru out of Google. Here's what it says. (MIT Tech Review) https://www.technologyreview.com/2020/12/04/1013294/google-ai-ethics-research-paper-forced-out-timnit-gebru/ (accessed May 14, 2024)

Neuralink's telepathy brain chip: How weird is it? (DW, Schwaller) https://www.dw.com/en/neuralinks-telepathy-brain-chip-how-weird-is-it/a-65227626 (accessed May 17, 2024)

Read the rest of DW's AI Lexicon:

A | B | C | D | E | F | G | H | I | J | K | L | M | N | O | P | Q | R | S | T | U | V | W | X | Y | Z

We're keen to hear your feedback. Suggest an entry by sending us a comment. And let us know if you feel we have missed something, got it wrong, and tell us whether our AI Lexicon has helped you understand the technology better.

Written and edited by: Zulfikar Abbany (za), Fred Schwaller (fs)