Artificial intelligence figures prominently in my EPSILON Sci-Fi Thriller series. My AI computer systems think like humans and perform tasks independently of human input. In Crimson Lucre, Robbie tends the hydroponic lettuce in the greenhouse. He surveils Pang Xianjing's garrison in Red Dragon.
What do we use AI for? AI is so ubiquitous we hardly recognize it anymore. Is that customer service rep on the phone a person or a bot? The results of a biopsy the doctor holds in her hand--did a technician or a program perform the diagnosis?
We are most familiar with AI in autonomous vehicle research and development. It executes tasks of vehicle control, sensor data analysis, trajectory prediction, and object identification. However, artificial intelligence does much more. In the domain of language and communication AI performs gesture, speech, handwriting and text recognition, and translation. AI's prowess at data mining facilitates quantum chemistry, facial identification, radar signal classification, 3D reconstruction and visualization, financial commodity trading, hydrology, ocean modeling, coastal engineering, and geomorphology. It diagnoses medical conditions, including several types of cancers. AI can distinguish highly invasive cancer cell lines from less invasive ones using only cell shape information.
In cybersecurity, it discriminates between legitimate activities and malicious ones. For example, it performs social network and e-mail spam filtering. AI systems classify Android malware, identify domains belonging to cybercriminals, find URLs posing a security risk, detect botnets, credit cards frauds and network intrusions. In short, we're building the Smith collection of programs in the matrix movies.
Will AI end the human race? I wouldn't be a bona fide Sci-Fi writer if I didn't ask this question, right? The novels, TV series and movies exploring this topic are too numerous to count. There are certain properties about artificial intelligence, and digital systems in general, that give us pause (more on this later). But the short answer is no, at least not with the AI technology we have today and in the foreseeable future. Here's why.
How does AI work? A computer uses artificial intelligence to think like a human and perform tasks on its own. Let's examine how AI works in real life.
A patient with lumps in a breast visits her doctor, who takes a punch biopsy and forwards it to a lab. A tech prepares a series of slides, creates digital images, and inputs them to abreast cancer diagnosis program. The software recognizes the image and analyzes it for the presence of cancer.
The program subjects the images to a number of algorithms--precise lists of instructions directing step by step actions in software-based routines. These algorithms are organized into a Neural Network. An NN consists of three layers, each of which may contain multiple algorithms.
An Input Layer searches the image for certain characteristics such as shape, color, texture, size (absolute and relative), etc. The IL forwards the constituent attributes to the Hidden Layer, which itself may have numerous layers, each with multiple algorithms. The Hidden Layer evaluates each property and assigns a statistical correlation to cancer based on its training. The properties and their statistical correlations are next sent to the Output Layer, which aggregates the probabilities and determines a final probability.
If the analysis outcome is 0%, the doctor will share the results with the patient and say, "You don't have cancer." If the result is a low number, say 20%, the doctor may choose to pursue further tests looking for genetic markers, or certain blood-borne proteins. If the probability is 95%, the doctor will evaluate appropriate treatment regimens.
How does the neural network identify cancer in the first place? Prior to deployment, it was subjected to a process called Machine Learning where the NN was shown millions of digital images of cancer. Its algorithms broke the images down into distinct properties and overtime established cancer cells display Property "X" 75% of the time, Property "Y" 80% of the time, and Property "Z" 95% of the time. It uses this analysis to construct a statistical model of a cancer cell. Once the model values stabilize, the program switches from training to deployment.
Once deployed, the model remains static. For any additional learning to take place the software must be taken out of deployment and fed the new or revised database--costly in terms of time and CPU usage.
What does AI do well? AI's main advantage over humans lies in its data analytics and pattern recognition. It can detect incredibly subtle patterns within large quantities of data. Our discussion on breast cancer diagnosis is a good example.
Artificial intelligence reduces human error. Its mind won't wander or make a mistake out of fatigue. It minimizes human risk. Coupled with robotics, it can perform in dangerous environments like a smoke-filled high-rise building. It's available 24/7. AI can operate in a factory setting without lunch or rest breaks or shift changes. It provides digital assistance and engagement--think Alexa or Siri. Businesses use AI for robotic process automation. It handles repetitive, rules-based tasks with great speed and accuracy.
Artificial intelligence develops new or improved inventions. Everything from designer molecules to novel engineering approaches to strength, durability, aerodynamics, weight savings, etc. Its pattern recognition in massive data sets offers analysis and insight, a powerful productivity tool.
It can offer unbiased decisions. Needless to say, this is a double-edged sword. AI is only as unbiased as its data set. In other words, artificial intelligence is completely amoral.
What does AI not do well? AI systems are rules-based and statistically correlated. They are highly constrained compared to human intelligence.
Artificial intelligence can't reason ethically. What's more, we don't know how to build an ethical AI , possessing amoral compass consistent with our own.
Theorists have proposed constructing ethical AI systems based on statistical models of how we act, but we humans aren't consistent ethically. We routinely implement our sense of right and wrong in a contradictory fashion. Rule-breaking is common when done for "the greater good," for a "higher purpose," or plain old self-interest. Generations of philosophers have grappled with human ethics without resolving the contradictions. There have been times when whole societies behaved unethically (Nazi Germany comes to mind), seemingly poor sources for AI ethical models.
Even objective artificial intelligence systems can become unethical if statistical anomalies exist within the data set. Facial recognition software, less accurate for brown-skinned faces, leads to false identifications in criminal cases, further burdening classes of people already struggling with disparities in housing, education, and economic opportunity. The amoral nature of AI requires thoughtful human supervision and intervention.
Artificial intelligence lacks commonsense. In other words, it can't apply learning from one domain to another situation or problem. Even minor changes in a task necessitates the system be entirely retrained.
Let's say we trained our AI robot to brew a cup of coffee, add just the right amount of cream and deliver it to the table in the breakfast nook. Now we ask our robot to bring a mug to our sunken living room at the base of a ramp. It would not keep the cup level to avoid spillage. If we provide it with smaller cups, our AI would not stop pouring when the liquid reaches the brim. Switched to porcelain china after we trained it using double-walled steel cups, we can expect a broken cup and spilled coffee.
We understand from life experience a fluid self-levels and won't maintain its position in a tipped vessel. We know a small cup has less capacity than a larger one, and we won't overfill it. We recognize a steel mug can endure more force and impact than a porcelain china cup and treat the fragile cup more delicately. Unless AI trains in specific rules for specific circumstances, it lacks the common sense of a human.
AI lacks dexterity, unable to learn continuously and adapt on the fly. This renders it incapable of dealing with unknowns or unstructured spaces.
This makes artificial intelligence impractical for exploring deep space in the absence of humans. Whether on Earth or on a distant planet, AI cannot apply information from one domain to another without going offline and retraining. In contrast, we can dynamically and smoothly incorporate continuous environmental input, adapting their behavior as they go. In other words, humans train and deploy in parallel and in real-time. Without close human supervision, AI quickly encounters circumstances beyond its training, catastrophically ending its mission.
Artificial intelligence can't understand cause and effect. With the right data, a machine learning model has no problem correlating when the wind blows a windmill turns. However, it's unable to distinguish if the wind causes the turbine to turn, or if the turning turbine causes the wind to blow.
Recalling the question posed at the beginning of this section, AI systems can't do the strategic planning necessary to end the human race. Even if some future artificial intelligence associates humans with "bad"--an ethical decision it's incapable of-- AI possesses no way to plan and execute the nuclear first strike to affect our demise. The ability to comprehend the requisite chain of cause and effect is simply not possible with digital architecture in use today or in theory. So, breathe a sigh of relief as you read the next great computer apocalypse novel, knowing it's pure fiction.
What will AI do in the next ten to fifteen years? Imagine a sweltering summer afternoon. For the past three hours you worked on a project in your garage. You're tired and hungry but don't want to make dinner.
"I know, I'll pick up something at the drive-through," you say to yourself.
Reaching inside the kitchen door you grab your Autonomous Vehicle key fob. The AV factory relied on AI to control its robotics. When a widget supply diminished, the program knew to order more widgets. The supplier, whose widgets were designed by AI, ramped up its 3-Dprinter fabrication. If the widgets were delayed, AV production at the factory was slowed based on self-generated projections. You toss up and re-grab your key fob, turn and walk back out into the garage.
As you approach your AV in the driveway, it recognizes your fob and unlocks the driver door. The onboard displays and interfaces wake up as you slip inside.
It's been a stressful week. You say, "De-stress music."
A soothing piano melody rises and falls to the accompaniment of rustling leaves. The piece spontaneously composed for you, based on your previous musical selections and preferences. You settle back into your seat, close your eyes and utter a contented sigh.
You're still hungry. "Go to McMegaBurger," you say. (Your future eating habits are obviously no better than mine at the present.)
Your hydrogen fuel cell electric AV backs out your driveway and onto your local street. Your car halts midway down the block when a ball bounces across its path. After waiting for the child to retrieve his ball and return to his yard the vehicle silently moves on.
Your AV stops at the intersection with a four-lane arterial, where it identifies a potential gap if one vehicle shifts over. Your AV communicates with the other car, which moves over one lane.
After entering the arterial your car adjusts its speed to match traffic. Scrolling through FaceAlbum to pass the time, you watch a video of a politician furtively stuffing money into his pockets. You aren't outraged by the content because a prominent label across the top of the image identifies it as a fake. Thanks to FaceAlbum's social network filtering and deepfake discrimination software, you weren't fooled.
After cruising a few minutes, your AV pulls into a two-way left-turn lane, waiting for a safe gap. One appears, but the car doesn't proceed. A mother with a stroller occupies the sidewalk in the driveway. It proceeds when the way is clear, and a new gap arrives. Once in the lot, the car yields control to you. You steer into the drive-through and rolldown your window.
A bot takes your order, flashing a text copy on the video screen. You confirm and pay by facing a camera. Facial recognition ties your face to your credit account. Transaction complete, you pull forward and pick up your meal at the automated take-out window.
The smell of burgers and fries fills your car as you tear into the bag.
"Take me home," you say before stuffing several fries into your mouth.
For Further Readinghttps://jlhancock.com/optogenetics/https://en.wikipedia.org/wiki/Artificial_neural_networkhttps://www.vaughn.edu/blog/artificial-intelligence-a-real-game-changer-in-the-aerospace-industry/https://www.simplilearn.com/advantages-and-disadvantages-of-artificial-intelligence-articlehttps://www.forbes.com/sites/forbestechcouncil/2020/12/16/what-ai-isnt-good-at/?sh=7de315965a7chttps://www.forbes.com/sites/robtoews/2021/06/01/what-artificial-intelligence-still-cant-do/?sh=1ce781d766f6https://bigthink.com/the-future/what-ai-cannot-do/#:~:text=AI%20cannot%20create%2C%20conceptualize%2C%20or,domains%20or%20apply%20common%20sense.
