Outsmarting AI. Brennan Pursell
Читать онлайн книгу.
That’s it, and it’s hardly new.
Computer hardware still relies on the same, simple architecture developed by John von Neumann in the 1940s. Hardware has input, output, and storage components. Software tells the hardware what to do and in what order. The computer just processes data. The central processing unit (CPU) still works with bits, on/off switches that indicate a 1 or a 0 (yes or no, true or false). The CPU works like someone in an old-fashioned mailroom picking up one number at a time from an in-basket, following the listed instructions, performing a calculation, and putting a number in an out-basket, over and over again. No real change here for eight decades, but vastly improved in speed, reliability, and efficiency. GPUs (graphics processing units) are specially purposed for images and can be used for AI processing, but they are still processors all the same.
A computer has no idea what it’s working on! It can’t explain anything. It neither understands nor learns like a human does. It’s just a stupid machine.
Human mental capability—call it intelligence, cognition, thought, perception, mind, feeling, whatever you like—is still a mystery to researchers and everyone else. The fact that no one can precisely define human mentality shows the most basic limit of our knowledge. Brains aren’t systems of linked, distinct, manufactured components, and they don’t act in long strings of 1s and 0s. Human thoughts are only very rarely like algorithms (step-by-step mathematical procedures).
A comparison between the human brain and a computer reveals few similarities. The table below sums up the vast differences between the two.
| HUMAN BRAIN | COMPUTER |
| 85 billion neurons (nerve cells) each connected to thousands of other nerve cells through its axons and dendrites (branches), with some 150 trillion synapses (connections). | Processors (CPU and GPU), memory (RAM), and storage are distinct hardware components. Size depends on how much you buy. The parts have few connections and cables for power and data. |
| 1 cubic millimeter of brain contains up to 100,000 neurons and about 1 billion synapses. | Transistors can get as small as a few atoms, but they don’t match the brain’s complexity. |
| Nerve cells are made up of organic molecules. Each is a living cell, and they come in many different shapes and sizes. | Bits are made of silicon and metal. They are all pretty much the same and equally dead. |
| Nerve cells take electro-chemical inputs and “fire” outputs that can use 100 different kinds of electro-chemical neurotransmitters. | The base unit is the bit, 1 or 0, on or off, in an electric circuit, a simple switch. In quantum computing, there is an “other” state. |
| Networks of neuron cells are constantly changing, on a daily basis. The brain “reprograms” itself, to some extent, according to need and ability. We don’t know how brains work to produce thoughts or manage memory. | AI “neural networks” are statistical software, distinct from the hardware, which cannot adjust itself unless programmed to do so, in limited ways. The software can be “trained” to self-adjust. |
| Electro-chemical processes for sensation, emotion, and consciousness are unknown. There is no center for decision, processing, or cohesive perception. There is no division between hardware and software. | Only the hardware performs mathematical processes according to the software. Hardware and software are completely distinct. |
| Brains are completely connected to bodies and can only work inside them. | Computers don’t need sensors. A keyboard and a monitor suffice for simple uses. |
| Brains mingle reason, sensation, emotion, perception, awareness, etc. | Computers do statistical classification and procedures. With digitized inputs and connected output devices, they can appear to simulate some human behaviors. |
| Brains understand, explain, and learn on their own. | Computers have no idea what they work on. They cannot learn as a human does, but the algorithms can be “trained” to issue certain outputs, given the input. |
So we don’t know what intelligence is or how the brain works or even how a single neuron works, and yet plenty of people claim that we can re-create, replicate, or replace a brain with a machine equivalent. The European Union allotted $1.3 billion to the Human Brain Project, an electronic simulation of 86 billion neurons and about a quadrillion synaptic connections in the vague hope that “emergent structures and behaviors” might turn up. We expect it to fail, grandly.[1]
But when it comes to mathematical calculations, computers have us beat! Processing speed is truly spectacular. In AI, algorithms work through oceans of data as fast as lightning to produce results that are used in turn to develop new algorithms that produce even more and better calculations, depending on your needs. Billions of calculations in thousandths of a second! Computers execute algorithms vastly faster and more reliably than a human ever could.
AI processes data. Without data—good, quality data, and lots of it—AI algorithms are useless. We can use AI tools to help make good decisions, based on data, for our organizations and ourselves. We make the decisions, not the machines.
Because AI is a brainless tool, it has to be managed, and governed, which means that we have to use it in accordance with the law. It is not enough to rely on the good intentions and lofty principles of AI ethics boards, committees, panels, partnerships, and professionals. The law has teeth, and it is a friend to good order, not an enemy. AI governance has to be done correctly, effectively, efficiently, and legally, for business success. For human success.
We’re not talking about the future. AI applications are already here, in every sector of the US economy. Across the globe, entrepreneurs and engineers work to improve these and develop new ones every day. COVID-19 neither slowed nor stopped them.
Joshua and I wrote this book to help you understand AI, be smarter than it is, to use it well, to control your data, and to govern your AI system, profitably and safely.
We Wrote This Book for People like You
This book is all about you and your work.
You are part of an organization of people—for profit, nonprofit, business, charity, government, NGO, what have you. And there are pressures on you to perform and compete. You can’t afford to ignore AI because it is an incredibly efficient mechanism that can be deployed and used effectively, if you design, deploy, and control it appropriately. You are already surrounded by it. You use it as a consumer whenever you search something on Google or buy anything from Amazon. It’s all over your smartphone. But it probably uses you more than you use it. Let’s turn that around!
You don’t need to be afraid of AI—it’s just a technology like any other—but you do need to watch carefully and learn how to use it. And you need to share this knowledge with your coworkers, superiors, employees, engineers, data scientists, suppliers, vendors, neighbors, and so on.
To outsmart AI, you do not need to be able to code all kinds of complicated algorithms. You do not need to be able to explain how GPUs help CPUs with data processing. But you do need to understand the differences among checklists, decision trees, and neural networks, so that you can make sure that AI tools are giving you the right information you require. You will need to maximize the quality of your data and make sure that your AI system does not churn out garbage or lead you to break the law.
We will help you. We don’t claim to be engineers or computer scientists. You don’t have to be either. We have had real-world success in designing, deploying, and selling AI and other advanced technologies. We have been teaching, writing, and helping people to work better for twenty years. Let’s ignore the hype, cut to the chase, and get to work.
How to Read This Book
Chapter 1 tears down seven widely repeated myths about AI. If we keep our heads, then we will master AI as we have all technologies that have emerged in human history. We are fools if we let it master us. In the same chapter, we present seven clear, key ideas about AI that will keep us straight on the path toward profitable AI implementation and proper AI governance.
Chapter 2 explains the science behind AI in plain English. I will show you what it can—and cannot—do for your organization.
Chapter 3 will