artificial intelligence and machine learning
Artificial Intelligence and Machine Learning Vocabulary:
- Machine learning
- Artificial intelligence
- Supervised learning
- Unsupervised learning
Up until recently, the idea of “giving” machines intelligence sounded like something straight out of an old science fiction novel. Well that seemingly borderline fantastical idea has become a reality in our world today. You’ve probably seen the terms “machine learning” (ML) and “artificial intelligence” (AI) quite often in the news these days. You may even have an idea of what these terms mean and represent. But while almost everyone has heard the acronyms “ML” and “AI” at some point, very few people have a deeper understanding of the technologies that these acronyms represent. It is crucial to have such an understanding, as ML and AI’s presence – and influence – in our society increases with each new day. In order to ensure that our future with these technologies is safe and that this tech is used for good, we must first learn about what it is, so let’s dive in!
What is it?
According to Dictionary.com, machine learning is “a branch of artificial intelligence in which a computer generates rules underlying or based on raw data that has been fed into it” and artificial intelligence is “the capacity of a computer to perform operations analogous to learning and decision making in humans”. It is important to note that ML and AI are not the same thing; ML is a branch of AI.
Key Ideas and Concepts – How Does Machine Learning Work?
Simply put, machine learning is a process that teaches computers something that comes naturally to us humans, which is learning from experience. ML algorithms use computing methods to “learn” information directly from data. (Note: an algorithm is a set of rules to be followed in problem-solving operations, especially by a computer)
ML uses two techniques – supervised learning and unsupervised learning. We will explore each of these techniques further in the upcoming section, but keep in mind this helpful analogy as we learn about these techniques: supervised learning is like learning with a teacher (the machine is taught what to do) and unsupervised learning is like learning without a teacher (the machine learns through observation).
A supervised learning algorithm involves using known input and output data; it uses this data to train a model to create reasonable predictions in response to new data. An example of supervised learning is having a data set of photos along with information about what is in each photo. You then train an ML model to recognize what is in a new set of photos (by using the photos + information data set to help the model learn).
Supervised learning uses classification and regression techniques to develop these predictive models.
- Classification techniques produce categorical (non-numerical) responses by classifying data into categories. For example, a classification algorithm can classify whether an email is real or spam.
- Regression techniques produce numerical responses. For example, a regression model can predict the temperature for the upcoming week in your town.
Remember that supervised learning = predictive models based on known data.
Unsupervised learning involves finding hidden patterns or structures in data. Here, datasets without known output are used. A common unsupervised learning technique is clustering. It helps to find patterns and groupings in data.
Remember that unsupervised learning = finding patterns/structures in unknown data.
Artificial Intelligence and machine learning have a great impact in our everyday lives and our society overall. Here are some real-world examples of how AI and ML used:
- Email spam filtering
- It may seem surprising, but AI helps keep your inbox safe! Did you know that Gmail successfully filters 99.9% of spam? It is able to do this because of machine learning. If Gmail was just given some rules such as ‘emails with phrases such as “credit card number”, “need money”, “Nigerian prince” should be marked spam’, it would not be so successful in filtering out spam because hackers could just update their messages and get around the filter. Instead, spam filters must continuously learn from many emails – and note what you may mark as spam – so that they can improve themselves and mark similar emails as spam in the future.
- Image recognition
- If you’ve ever uploaded a photo to Facebook, you’ll know that AI immediately gets to work on your post. Facebook’s software can recognize faces in photos that you upload and they can even suggest who’s faces they may be based on your ‘friends’ list and photos that your friends have uploaded.
- Tumor classification
- ML models can help pathologists distinguish different types of lung cancers in scans/images. These models would use classification techniques to do so.
- Self-driving cars
- In order for a car to be self-driving, the car needs to be able to see and perceive its surroundings. To do this, it uses image classification.
Why we need to be careful:
As you just read above, ML and AI can be applied in amazing ways and can be used for good. At the same time, these technologies can take a wrong turn if they are used irresponsibly, and if there is a lack of diversity in their field. The following is an excerpt that I wrote in an essay titled “The AIpocalypse: Why Humanity Needs To Take Action On AI Regulation”:
“A rather unexpected risk that AI comes with is its inherent bias. People believe that one of the biggest advantages about giving a machine decision-making abilities is creating an objective thinking counterpart to humans; however, since humans are the ones programming it, we are programming our biases right into it. This bias could very well alter the course of someone’s life – and not in a positive way. Recently, Amazon began utilizing an artificial intelligence tool to help with hiring. However, they soon had to stop using it because it was biased against women. It downgraded resumes containing the word “women’s” and it had filtered out candidates that had attended women’s only colleges. Just imagine yourself as a woman in this situation – you have worked incredibly hard for twenty-something years of your life (following your passions, pushing yourself, learning as much as you can), and you go to apply for your dream job, only to find that your fate is to be determined by a machine that is trained to discriminate against you. That’s a facet of AI that is truly flawed. But it is a facet of AI that has a potential solution – and we can and must work towards it.
It is a known fact that regardless of the amount of risks that AI poses, humanity will not stop charging towards an AI-filled future. So the most viable path to a solution that we can take involves not attempting to stop the spread of AI, but rather taking steps towards reducing its negative impact in the future. With the rate at which AI is advancing, we must begin taking steps to mitigate its risks starting now. One way to do this is to give minorities in tech – such as women – the opportunity to program AI. Currently, AI is developing a bias against minorities because there are a lack of minority groups programming it. Thus, in order to ensure that AI is more objective, we, as a society, must encourage the diversification of AI programmers.”
Outline of the topics touched upon in this article:
- What ML and AI are
- How ML works
- Supervised and unsupervised learning
- Classification, regression, and clustering techniques
- Common uses of ML and AI
- Why we need to be careful with this ML and AI
Fascinated with the topics of machine learning and artificial intelligence?. Read more below if you are interested!
AI and ML Vocab Defined:
- Machine learning – a branch of artificial intelligence in which a computer generates rules underlying or based on raw data that has been fed into it
- Artificial intelligence – the capacity of a computer to perform operations analogous to learning and decision making in humans
- Algorithm – a process or set of rules to be followed in calculations or other problem-solving operations, especially by a computer
- Supervised learning – involves using known input and output data; it uses this data to train a model to create reasonable predictions in response to new data
- Classification – problem of identifying to which of a set of categories (sub-populations) a new observation belongs
- Regression – models a target prediction value based on independent variables
- Unsupervised learning – involves finding hidden patterns or structures in data and using training data with unknown outputs
- Clustering – machine learning technique that involves the grouping of data points