Читать книгу Cyberphysical Smart Cities Infrastructures. Optimal Operation and Intelligent Decision Making онлайн

25 страница из 94

In a traditional sense, cybersecurity focuses largely on the protection of data via access controls, encryption, and other methods. The idea is that when one machine is sending data, only those who should be seeing the data get to, and it does not get modified in between transmission and receiving. To put things into a physical perspective, think of mailing a letter. If one sends a postcard to someone, there is nothing stopping every single person from reading what they wrote on the card. Now, if instead that letter was put in an envelope, it would prevent outsiders from freely reading what you wrote. To take it one step further, what if there was a concern about someone opening the letter and reading it. Perhaps in this case, the letter is placed in a box that is locked and sent to the receiver who has the key (this is how encryption works). The more important and private the letter, the more protection placed on it, ensuring that only the receiver is the one to read it.

This closed‐off system poses a problem for AI cybersecurity. Gideon explains that for AI to function as they are designed, they need access to steady streams of data [5]. Going back to the letter analogy, if you bring AI into the mix, then AI algorithms would need a special way to get inside the box, be able to read the letter, and extract information from it. If someone were to break in and learn about these vulnerabilities, they could potentially exploit them by altering the data or causing the algorithm to behave other than intended.