Logic and reason are so important in how we relate to our world. Reacting in emotionalism will only lead to error and regretful mistakes. Acting with emotion alone, people can harm others or harm corporations, thinking they are hitting back against a problem but in reality are creating a new problem. I want to discuss the topic of AI from a reasonable and logical point of reference, where various outrages are handled and addressed.
Recently, I witnessed a hive mind of people attacking the concept of AI with a misplaced anger to a product. The product was Topaz’s Photo AI tool. The tool itself is interesting. It combines several of their products that supposedly utilize AI to denoise an image, upscale an image, remove unwanted elements and resharpen blurry photos. It works very well at achieving these things, and I’ve used Topaz’s Gigapixel for about a year, to upscale images 8x-10x without pixelation.
So why are people angry with Topaz and their products? Because they use the word “AI” in their promotional material. The comments of the promotional ad were very hostile to the use of AI in photography. I believe the anger is misplaced and I want to address AI in general with some point and counter points to what I noticed in the comments.
At the bottom of this article are common Arguments raised and responses to them.
What exactly is AI?
Artificial Intelligence is often misunderstood. AI became a buzzword very quickly as it was a way to capture the attention of people interested in the technology. But what exactly is AI?
Years ago I read an article that talked about the use of AI in the 1950’s. In fact much of the modern work of AI today is based on early concepts from the 1950’s and 60’s. At that point in time, the direction of computing was at a fork: algorithmic programming or artificial intelligence. With the restriction of large data sets and computing power, algorithmic programming won out as the direction of computing for sixty some years.
Algorithmic vs. AI
Algorithmic programming is a formula driven approach to software design. An algorithm is a set of rules in a process that performs a specific function. Pre-computers, mathematical concepts were based on algorithmic thinking. Logic and formula are utilized to take input and return a given output. Input outside the scope of what was programmed, is ignored or errors out. In other words, with this type of software, the designer accounts for each type of interaction. Adobe’s classic Photoshop is an example of algorithmic programming. The software has sliders and buttons and filters that perform programmed (pre-conceived and defined) rules. Adjusting the contrast by moving a slider, increases or decreases the amount of contrast. The User Interface (UI/UX) allows for the formula to add or subtract a level of contrast in that specific algorithm.
AI (artificial intelligence) uses an underlying framework of statistical inference and analysis that analyzes large data sets. From this data and parameters set, an AI application makes a decision. In the AI subset of Machine Learning, for example, decisions are based on Linear Algebra. In machine learning data points are met with a “best fit” line that runs through the center of the data. A machine learning system will make a decision based on the data that is that best fit. The decision is later analyzed to see if it is correct or false and a better fit is attempted. This is a continuous process of “learning.” AI is a larger than machine learning, which is a subset of AI. However, the concept of continuous fit gives a simple and clear idea of how a machine can “learn.” AI requires lots of data. The more digital data it can ingest to analyze for a problem, the better result it will have towards the desired goal.
Where AI differs from programmed logic, is that AI uses a trained model to rapidly predict a best course or best decision. A trained model is where the AI application has been given datasets that were trained to accept specific patterns or choices that had favorable outcomes. The key here is that this model isn’t requiring human guidance. There isn’t a programmed “do this, else do that,” but instead an AI program relies on input coming in, looks for patterns, uses the trained data (what worked in the past) and formulates a best decision. Human effort here was back during the training phase. Now the machine is making the output, without human programming. In other words there is more flexibility in what a machine will decide to do.
Examples
If one were to program how a car will park itself, it would have a set of predefined rules. Rules such as using a camera to identify a parking area. Logic to determine “is this a head in parking space or parallel parking space” and then run a set of pre-programmed rules to start the parking mechanism. Rules like slowly move forward into the head-in parking space. Apply brake pressure. Come to a complete stop.
Unfortunately pre-programming can’t consider a variety of issues that may randomly occur. What if another car attempts to park in the same space at the same time? What if someone walks in front of the car, behind the car, or what if [any numerous event] occurs? Whatever you can conceive, you can program for. But when dealing with the physical world, the chaos of reality is infinite. Instead of relying on a set of rules to park the car, AI would rely on a trained model. The model would have data from thousands of parking events, and the concept of parking correctly in each incident. It would know how to deal with different weather conditions, environment changes, interaction with people or objects. With those data scenarios, having fully been trained, it would then work out solutions in real time.
“What if it’s decision is bad,” someone will ask. But to answer that, human decisions are also fallible. A choice to hit a dog in the road in front of you, or slam the breaks and be rear ended, is a tough one. A human may choose to kill the dog, or swerve, or several other options. All can be “wrong” when seen in the light of revisionism. Putting obvious errors aside, how can we judge AI when we ourselves would struggle with the same decision?
Concerns of AI
When Chat-GPT and Midjourney hit the stage with their creative AI offerings, it pretty much shocked the world. AI was no longer just used in cancer research but was now something everyone could play with. People began pumping out millions of images created with AI. Chat GPT was able to write essays, and other tools like AIVA was producing music in several different genres. Creatives became concerned very rapidly over two main issues:
- Did the AI system steal from copyrighted material to “train” it
- Will AI replace me
Copyright and Training
Several times this has been raised that an AI utility creates content using a phrase or image from a copyrighted source. If copyrighted content was used to train AI, then we have an inherent problem in the output. Every so often the AI will generate output that is copyrighted. These situations occurred early on and were met with court cases.
It’s challenging to prove that AI has been trained on copyrighted data. There’s no need to use copyrighted data as the public domain is filled with millions of data points. Most image sharing sites likely have a clause about their rights to any image hosted/used on their platform. That may include the use of the image for training purposes.
While this is an issue, hopefully it will be dealt with so that copyright concerns are not a problem. One step to achieve this, might be to run all generated content through a plagiarism checker (which is also likely AI) to validate the generated output is without issue.
AI replacement
AI will replace jobs. This is a fact. As every technological advancement replaced human roles, so AI will continue the trend. Areas that will quickly be impacted are photo stock, artwork, art prints, concept art, music and writing. Further impact will likely occur with 3D artists, novelists (long form text generation), and programmers.
That doesn’t mean all jobs are lost. But those industries will be impacted.
Imagine a small company that struggles to make profit. They could hire 4 artists to design their logos, posters, social media campaigns or they could hire 1 or 2 and those people would be instructed to use AI to leverage the need for workers. The artist becomes more of an Art Director, handing off their vision to a training model that rapidly returns designs. This process is nudged along by the staff artist until the desired goal is achieved and they likely will make small changes for their desired goal.
AI can not fully replace all artists, because it doesn’t understand emotional impact and it currently lacks consistency. Each generated image is a unique concept, without the integrated project concept. What color should be used for this audience? What palette should be used? What images are pleasing to the audience/client? That still requires human effort. For a machine to make those choices, it would have to understand the likes and wishes of an audience and then create images that relate, and be able to constantly update with customer likes. It’s possible this will happen in the distant future. For now though, I think an artist will transition into an art director, passing a vision off to a machine (instead of a human artist).
Music and writing are other impacted jobs that again will have a point person who passes off desired goals to an AI that generates the content. Human ingenuity is still required as the concepts, and goals of what needs to be written will be required. AI still generates generalized content that lacks specificity. In time that challenge will probably be resolved and it will produce specific, detailed analysis in music and writing content.
Laws and regulations to protect jobs
Personally, I think the short term will have many nations fighting AI with regulations requiring the use of human capital. Long term is a different story. Some nations will push on with the least cost (AI) and achieve rapid prototyping. This will become a pain point for other competitive nations who will ultimately (I think) bend to the competitive need and remove all regulation in regards to AI and human capital.
Human protest
Currently people are protesting the use of AI in creative content production. When a movie is spotted using AI in its poster design, or an album cover is spotted to have been generated from AI, boycotts and human reaction occur. It’s an attempt to pressure the industry into focusing on using human capital.
For large studios, cost of human capital isn’t an issue. What they will likely want with AI is rapid prototyping. I think studios will align with human capital.
However, small to medium sized companies who struggle for a competitive edge, will find it difficult to compete with human capital alone and will likely leverage AI to some degree, regardless of protest.
Benefits of AI
AI has made impact in medical and research fields for quite some time. Future development will require more involvement of AI. AI will be at the cornerstone of finding cures, achieving cost effective solutions to problems, and be able to increase rapid prototyping with less loss.
Commercial uses
Everything that collects or reads data will likely have an AI component at some point.
My sprinkler system uses AI to decide when to water the grass. Using weather data, it compiles the best watering times for my specific lawn type with the weather patterns for the week.
Smart homes are using AI to help lower costs on electricity. By modifying the flow of electricity AI can auto dim lights during peak charge hours, or can set the best temperatures for the AC/Furnaces in a cost effective way.
Garage doors can be setup with AI to facial recognize and / or use voice recognition to open or close.
There are so many applications of AI in commercial projects.
Technology uses
AI in technology will rapidly increase our capabilities. As sophistication of technology increases, human capabilities to monitor and adjust to rapidly changing data becomes impossible. This is where AI is leveraged.
AI used in vulnerability management is a new frontier, but will become a staple as threat actors utilize AI against targets. NGFW’s will likely start using AI as SIEMs and SOARs are doing it already. Elastic has AI built in to monitor network traffic and identify potential suspicious or threat traffic based on patterns of behavior.
AI is also going to be used in more and more safety features in transportation. Buses, cars, trains and maybe even planes will have some sort of AI guidance or mitigation measures. This will benefit human safety as well as enable cost saving measures (such as a mode to allow AI to adjust speed to provide the best gas economy).
Financial technology has been using AI to better price homes at market demand. This will likely improve non-professional investors with investment ideas.
Medical and research uses
AI was working in the medical and research fields early on and it continues to make major wins. AI has found new indicators for cancer in patients. By analyzing data, it uses pattern finding to discover predictors towards cancer or other diseases.
Using data from thousands of hospital patients, AI can be trained to find statistical relevant issues that could appear in a person’s life.
Scientific research will find more gains using AI for pattern recognition, allow for rapid response to fast changing systems like fuel injection.
Life expectancy and human augmentation will increase with the use of AI.
The Arts uses
In the arts, AI helps us with prototyping, coming up with new ideas and concepts that we can work into our own projects. AI can benefit music by creating midi chords and melodies to overlap our own efforts.
In the visual arts AI can be used to create prototypes, concepts and photo real results. Art can be generated to a stylized look (abstract, impressionist) or photo real.
Language Learning
As AI can ingest lots of data, it can use data on any language to teach the language through conversation.
Psychological uses
AI trained on therapy models can be used as a constant companion to help leverage times the therapist is not available. This can help people air their thoughts and feelings without drawing down on the therapists time.
Cost mitigation
AI has lowered the overall cost of production. Smaller companies will save money by switching from Adobe stock subscriptions, to generating photo real images using Midjourney.
Smaller companies and home users will benefit from the lower costs to grammar check, re-write essays, give insights into research, and so on.
Not All AI is the Same
When I see irrational hatred of AI, it tends to consider one aspect of AI they are frustrated by (usually art generation), and they misplace that anger to all other forms of AI.
As previously mentioned, there is AI used in cars, in sprinkler systems, in photo editing, in medical research and this form of AI isn’t impacting people’s income or copyright.
Arguments and Counter Arguments
I’ve come across several arguments and have provided counter arguments below.
Specific to Photo Editors, I’ve seen these arguments:
“This AI Tool is evil because all AI is evil”
I’ve seen people attack photo editing tools (like Topaz Photo AI, or Luminar Neo) chiding them for using AI. The AI used in these tools is to modify and correct their own photos. These tools do things like increasing the size of an image without pixelation, sharpen blurry shots, remove dust or digital noise and unwanted items.
They are fixated on AI taking over, taking their jobs, etc. But the tools they often attack are not taking away someone’s job. Tools by Topaz and Luminar are not taking away someone’s job, as when’s the last time anyone sent their vacation photos to be edited by someone else? That’s usually a personal choice we make and do on our own.
These arguments are often straw man fallacies.
“Removing objects from an image is taking away the Truth of photography”
One anti-AI person threw this argument at me. I showed how removing objects from an image is something that goes back to the film days. Pre-AI, digital, we used tools like Photoshop to clone over areas we don’t want. A bird in the sky can be cloned out by replacing the bird with other parts of the sky. People could be cloned out or painted over. Yet no one in the past 25 years screamed, “shame on you for removing objects from your own images, with Photoshop!”
Since this argument is only now appearing as a response to AI tools, I think it is a problem of false equivalence as it really isn’t against “truth in photography” at all, since these objects were absent for the past 25 years.
“I don’t want my image to end up looking like everyone else’s”
I can understand this idea when using templates. But this problem isn’t an AI problem. Pre-AI people used Instagram, Lightroom and other tools to sensationalize an image to a certain look. Phones do this a lot. Phone cameras often oversaturate or sharpen or remove noise programatically. So again, this is a straw man fallacy.
“AI is theft”
As with human artwork and design, there are always copyright concerns. AI is no different. When we learn from different artists, their styles may influence our own and sometimes it may be such an influence that the overlap becomes an issue of copyright law. We learn how to do things in a way very similar to AI. We take in data and from that data reproduce a desired output. If the data is copyrighted, sometimes this is a problem.
The statement that “AI is theft,” only accounts for one use case of AI (in art generation) and not the much larger umbrella of AI. AI is used in smart home tech, sprinkler systems, medical devices, scientific research, self-driving cars and in these cases it is not stealing. So we can’t logically say, “AI is theft” because that infers “All AI is theft.”
Regarding the specific use case of AI in image generation, copyrighted material may appear in art generation. These must be considered in their respective cases, just like with human art creation.
“AI is a tool for non creatives”
Plenty of creative people have started using AI in their projects. From music production, to visual art, AI is used to prototype ideas, or modify their own work to further enhance the result.
Of course this is a very subjective opinion/statement/argument. I can’t state emphatically it’s wrong, but more that it is too generalized and misses that many creatives use AI in their work.