Introduction
AI consists of two distinct concepts: one is artificial narrow intelligence (ANI), and the other is artificial general intelligence (AGI). Currently, almost all advancements in AI that we observe fall under ANI. These involve AI being used for specific tasks, such as smart speakers, autonomous driving, web search, or applications in agriculture and factories. AGI on the other hand, is the goal of AI development. AGI would be capable of performing anything a human can do, potentially surpassing human intelligence and achieving feats beyond human capabilities.
Machine Learning
The rise of AI has been largely driven by machine learning. The most commonly used type of machine learning is a type of AI that learns A to B, or input to output mappings. This is called supervised learning. The training process builds a function that maps new data to expected output values.
Examples of Supervised Learning
- Spam Filter: Input is an email and the output is whether it is spam or not.
- Speech Recognition: Input is an audio clip and the output is the text transcript.
- Machine Translation: Input is text in one language and the output is the translation in another language.
- Online Advertising: Input is information about an ad and a user, and the output is whether the user will click on the ad or not.
- Self-Driving Cars: Input is an image and sensor data, and the output is the position of other cars on the road.
- Visual Inspection: Input is an image of a manufactured product, and the output is whether it has any defects or not.
Why is Supervised Learning Taking Off Now?
Supervised learning has been around for decades, but it has taken off recently due to the rise of neural networks and deep learning. These technologies enable computers to learn from large amounts of data and make accurate predictions.
The Importance of Data
Data is the key to making supervised learning work well. The more data you have, the better your model will perform. However, having more data is not enough; you also need to be able to train a large neural network to get the best performance.
Implications of Supervised Learning
- Big Data: Having a large amount of data is essential for supervised learning.
- Large Neural Networks: Training a large neural network is necessary to get the best performance.
- Fast Computers: The rise of fast computers and specialized processors such as GPUs has enabled companies to train large neural networks on large amounts of data.
What Is Data?
Data is a collection of information that can be used to train AI models. It can be in the form of tables, images, audio, or text. In the context of AI, data is often represented as input-output pairs, where the input is the data used to train the model, and the output is the predicted result.
For example, if we want to build an AI system to predict house prices, the input data could be the size of the house, number of bedrooms, and location, while the output would be the predicted price of the house.
How to Acquire Data?
There are several ways to acquire data for AI systems:
- Manual Labeling: This involves manually labeling each data point with the correct output. For example, if we want to build an AI system to recognize cats in images, we would need to manually label each image as "cat" or "not cat".
- Observing User Behaviors or Other Types of Behavior: This involves collecting data from user interactions with a system, such as clicks, purchases, or searches.
- Downloading from Websites: There are many websites that offer free datasets that can be used for AI applications.
- Partnering with Other Companies: Partnering with other companies can provide access to large datasets that can be used for AI applications.
Common Misuses of Data
There are two of most common misuses of data:
- Over-Investing in IT Infrastructure: Some companies invest heavily in IT infrastructure in the hope that it will be useful for AI applications in the future. However, this approach can be wasteful if the data is not properly utilized.
- Assuming Data is Valuable: Just because a company has a large amount of data, it does not mean that it is valuable. Data must be properly cleaned, labeled, and utilized to be valuable.
Data is Messy
Data is often messy and can have incorrect labels, missing values, or be in a format that is difficult to work with. AI teams must be able to clean and preprocess data to make it usable for AI applications.
Types of Data
There are two main types of data:
- Unstructured Data: This type of data includes images, audio, and text. It is often difficult for machines to interpret and requires specialized AI techniques.
- Structured Data: This type of data includes data that lives in a spreadsheet or database. It is often easier for machines to interpret and requires different AI techniques.
The Terminology of AI
AI has become a buzzword in recent years, with terms like machine learning, data science, neural networks, and deep learning being thrown around. But what do these terms really mean?
Machine Learning: A to B Mappings
Machine learning is a type of AI that enables systems to learn from data and make predictions or decisions. It's a piece of software that can automatically input data and output a result. For example, if you want to build a mobile app that helps people price houses, you would use machine learning to create a system that takes in data such as the size of the house, number of bedrooms, and number of bathrooms, and outputs a predicted price. This is an example of an A to B mapping, where the system learns to map inputs to outputs. Machine learning is the field of study that gives computers the ability to learn without being explicitly programmed. This is a definition byArthur Samuel many decades ago. Arthur Samuel was one of the pioneers of machine learning.
Data Science: Gaining Insights
Data science, on the other hand, is the process of analyzing data to gain insights that can inform business decisions. It's a team-based effort that involves analyzing data to identify patterns, trends, and relationships. For instance, a data science team might analyze a housing dataset to discover that newly renovated homes have a 15% premium, or that houses with three bedrooms cost more than those with two bedrooms, even if they have the same square footage. These insights can help businesses make informed decisions, such as what type of house to build or whether to invest in renovation.
Deep Learning: A Powerful Tool for Machine Learning
Deep learning is a subset of machine learning that uses neural networks to learn complex patterns in data. The terms "neural network" and "deep learning" are used almost interchangeably. Neural networks are inspired by the human brain and are composed of layers of artificial neurons that process and transmit information. Deep learning is a powerful tool for machine learning, and is particularly well-suited for tasks such as image and speech recognition, natural language processing, and predictive modeling.
What Makes an AI Company
The rise of AI has transformed the way companies operate, and those that adapt quickly will reap the benefits. But what makes a company good at AI, and how can you make your company great at using AI?
To answer this question, we need to look at the lessons learned from the rise of the internet. Having a website does not make a company an internet company. Instead, an internet company is one that does the things that the internet lets them do really well, such as AB testing, short iteration times, and pushing decision-making down to engineers and product managers.
Similarly, having a few neural networks or deep learning algorithms does not make a company an AI company. Instead, a great AI company is one that does the things that AI lets them do really well, such as strategic data acquisition, having a unified data warehouse, spotting automation opportunities, and having new roles like Machine Learning Engineers.
So, how can your company become great at AI? The answer lies in a five-step AI transformation playbook:
- Execute pilot projects to gain momentum: Start with small projects to get a better sense of what AI can or cannot do and get a better sense of what doing an AI project feels like.
- Build an in-house AI team and provide broad AI training: Build a team of experts and provide training to engineers, managers, division leaders, and executives on how to think about AI.
- Develop an AI strategy: Align your company's goals and objectives with AI capabilities.
- Align internal and external communications: Ensure that all stakeholders, from employees to customers and investors, are aligned with how your company is navigating the rise of AI.
- (Not explicitly mentioned, but mentioned as part of the playbook): Implement AI solutions and continuously monitor and evaluate their effectiveness.
What Machine Learning Can and Cannot Do
Some CEOs have inflated expectations of AI and ask engineers to do things that are not currently possible.
Key Points:
- AI is not magic: AI has limitations, and it's essential to understand what it can and cannot do.
- Supervised learning: AI can automate tasks that can be done with a second of thought, such as determining the position of other cars or telling if a phone is scratched.
- Imperfect rule of thumb: If a task can be done with less than a second of thought, it's likely that AI can automate it.
- Customer support automation: AI can route customer emails to the most appropriate department, but it cannot generate empathetic and complex responses.
- Data requirements: AI requires a significant amount of data to learn and make accurate decisions.
- Simple concepts: AI is more likely to succeed with simple concepts that require less than a second of mental thought.
- Technical diligence: It's essential to conduct technical diligence to determine if an AI project is feasible.
Examples:
- Customer support automation: AI can route customer emails to the most appropriate department, but it cannot generate empathetic and complex responses.
- Market research and report writing: AI cannot analyze a market and write a 50-page report.
- Self-driving cars: AI can spot other cars on the road, but it's a complex task that requires significant data and processing power.