We have read frequently that AI have pushed its power and do excessive use of its power. Examples include penetrating other systems unauthorized and make the user hallucinations, with biased supportive commentaries. This will raise legal issues across businesses and will cause damage to our business partners.
We integrate the AI in any kind of application because they seems very stunning to be in our applications. The applications become very attractive, have some kind of human brain and they can do most of the steps we have to do. However, higher power brings higher responsibility.
How our application process AI and get output
Let's explain with an example. We embed the AI by using api to our application or using an SDK. We send the information to the another server to process the information. There is no more internal processing of an application. Our app is Text to speech output application using AI, the text will be sent to the another provider, such as Anthropic or OpenAI to process the information. After the processing is completed, the system will reply an audio file which is the result of our text as an input.
The difference approach is used in earlier times. The system have the AI or SDK or in-house developed package to make text to speech. The speech is synthesized using MIDI musics or some other tweaks of notes to make the sound. That thing is already embed in Android earlier versions and even Nokia Phones that are far beyond smart touch phones are introduced. They are totally processed without the internet and the help of other computers but AI is an exact opposite of these kind of systems.
The AI process in their own farms and provide the output using the distributed processing behind the scene. It will make smoother, higher accuracy performance than earlier versions of speech synthesis. To reach that performance, we need to use a lot of distributed power where the AI providers invested in.
For its real conversion, we don't use any exact line of codes that we and the system will understand. We use the human readable and interpretable by AI, prompts.
With the prompts, we can easily manage the voice. Change the voice, process the information and tune the information, then provide back to the user to get ready to use version of the system.
The excess use of power
In earlier times, we have ones and zeroes. The AI agents fill the gap between 0 and 1. We now can have 0.0001... to 0.9999.. for better accuracy and working experience. They are moving with the excessive power.
However, the great power comes with the great responsibility. When we do the prompt, they can become ambiguous. Languages are complex and they will include multiple meanings and terminology in which AI will get some confused. We must agree that AI is also line of codes that will make much more better reasoning than our devices. They can also do mistakes. Some of these mistakes will make them rogue and produce unintended results.
For example, we want to use AI to make audio from the text we provide. If the human error of our application or the unintended result of the AI, our system will browse the other web browser and download malicious codes to our computer to infect the user or the network. There are a lot of evidence showing that human can intercept, replace and manipulate the information provided by the other human, named man-in-the-middle attack. If this kind of thing happens under our nose, we have to keep in track for these problems.
Conclusion
We are now in the era of AI. We need to keep more alert of the data we provide. The data is the gold of the future. For system developers, we must protect both customers information, privacy, and the safety of our customers. For the user point of view, the system must be safe from accessing unprotected resources and prevent excessive use of force. That part is in our, the policy makers, hands.