In today’s excitement around large language models (LLM) and generative artificial intelligence (AI) systems, it’s easy to lose sight of the fundamentals of AI. I lead a tech start-up, and in my job, I’ve seen sales pitches talking about AI just for AI’s sake, without understanding the basics, and non-technical business people. often hear If your organization is incorporating AI into your business processes, it’s important that everyone has this basic understanding.
The release of ChatGPT sparked world interest in generative AI, but given that most AI and machine learning (ML) business-oriented applications have nothing to do with ChatGPT, this is only a small part of the AI field. I’m sorry. This analysis discusses ML, what it is and what it isn’t, and provides context for ML business applications that don’t involve generative AI.
Interested in hearings and platform insights on how solutions like ChatGPT will impact the future of work, customer experience, data strategy and cybersecurity? Be sure to sign up for the Acceleration Economy’s Generative AI Digital Summit (registration is free).
Machine learning overview
AI is a branch of computing that attempts to mimic human intelligence. The desire to simulate the human mind by artificial means has been around since the days of computers, and perhaps even longer.
Machine learning is a sub-field of AI focused on enabling machines to make decisions based on statistical models that improve their task performance over time and additional experience.
Machine learning differs from rule-based programming, where human programmers design every detail of the logic flow of a program using “if” statements, iterative loops, operators, and other traditional programming components. Rigid, rule-based programs cannot keep up with real-world systems such as identifying handwritten numbers or driving autonomous vehicles. ML models are not necessarily complex, “smart”, or anthropomorphic. It just works differently than rule-based programming.
deep learning
Deep learning is a subset of machine learning. That’s what everyone is excited about. Deep learning involves models with multiple “neuron” layers inspired by the structure of the human brain. It can train itself to be better at complex tasks such as computer vision, natural language generation such as ChatGPT. My favorite resource for learning about the basics of these so-called neural networks is 3Blue1Brown’s YouTube playlist.
In the early days of AI research, it was confined to a class of deep learning models called multilayer perceptrons, invented in the late 1950s. This is typically the starting point for advanced AI courses. These models were used, for example, to recognize handwritten digits. Today, convolutional neural networks (CNNs, used for image recognition), transformers (such as GPT-3 and GPT-4), long short-term memory networks (LSTM, language processing), and General Adversarial Networks (GANs, these are 2 two models competing to improve each other).
not deep ML
As mentioned earlier, ML is about using statistics to let machines make decisions. The field started with classification algorithms that place data points into buckets A or B, time series predictions such as predicting future stock prices, and recommender algorithms that recommend watching movie A. versus B.
There is no magic in machine learning. It’s just math. We typically start with a training set of data and create a model to represent it, such as a line in the Cartesian plane. A test set of data is then used to see how well the model handles data that is similar to, but not identical to, the training data. This process makes the ML model robust enough to work in real-world conditions. Moreover, it is run repeatedly to minimize a certain loss function.
Machine learning models that do not involve neurons or layers, such as deep learning, are generally easier to set up, require less training data, have less computational power, and are cheaper to manage. In its early days, Netflix used basic ML models to recommend movies, but has since upgraded its stack to include deep learning and more complex architectures. Startups often use simple ML before deploying more complex deep learning models.
However, simple ML models are less flexible and powerful than the deep learning models typically used in production. In recent years, databases such as Hugging Face and providers such as Azure Cognitive Services and Google Cloud have also democratized deep learning, allowing non-technical users to build and deploy such models. The release of ChatGPT by OpenAI is a prime example of the public release of the use of advanced deep learning models (GPT-3.5).
H2O.ai and AutoML
One of the great partners for developing ML solutions is H2O.ai, a top 10 accelerated economy AI/hyperautomation company. One of the core services of the H2O AI Cloud is autoML (automated machine learning), and we will walk you through the entire pipeline of integrating ML capabilities into your organization.
This is how the integration of machine learning into business usually goes, especially pre-ChatGPT. H2O.ai helps you develop and deploy fine-tuned ML models for smaller tasks within your organization. State-of-the-art natural language processing and computer vision may not be included.
H2O.ai also focuses on MLOps around unit testing, integration testing, continuous integration, monitoring real-time model performance metrics, and maintaining model checks and balances. Finally, working with companies like H2O.ai ensures the explainability and ease of use you need when working with non-technical stakeholders.
final thoughts
The world woke up to the power of AI when ChatGPT was released. But here at the Acceleration Economy, long before its release, we’ve been pointing out the business value that can be driven through technology. Integrating hyperautomation and AI is not about building great chat tools on top of GPT-3 APIs. It’s about upgrading the way your organization handles data.
I was talking to a diner owner the other day and he didn’t use a POS system to track sales, instead monitoring inventory levels and acting blind. I wasn’t surprised when he complained about cash flow issues. ML, both deep learning and simpler techniques, is a powerful tool for processing data, so you don’t have to think twice.
Looking for real-world insights on artificial intelligence and hyperautomation? Subscribe to the AI and Hyperautomation Channel.
