AI-generated painting—-creating or stealing?

17

October

2022

No ratings yet.

After AI has successively conquered problems such as chess, speech recognition, and text translation, AI-generated paintings are nothing new. The ability of AI to paint has begun to appear many years ago, and many technology companies are currently deploying in this regard. As early as a few years ago, painters used Paintschainer to colorize the line art drawn by themselves through the computer, and the effect achieved can already catch up with some relatively simple artificial coloring works (Ren, 2019). In addition, through the artificial intelligence image generation scheme of GauGAN provided by NVIDIA, the painter only needs to simply smear a few strokes and roughly outline with different color blocks and lines, and GauGAN can guess the painter’s intention and actively draw out An extremely realistic photo (Park, 2019). GauGAN 2 uses 10 million images for training, and artists can use it to generate not only realistic landscapes, but also imaginative artistic scenes, which have surpassed many people’s traditional impression of computer illustration art.

In March 2016, with AlphaGo defeating the Go world champion Lee Sedol in the Go competition, AI not only won the attention of the outside world but also ushered in a period of rapid development of its industrialization. In fact, behind the rapid development of AI, there is an inseparable relationship with the development of three types of technologies: graphics processing unit (GPU), deep learning, and big data.


Take AI drawing as an example. In fact, the most fundamental basic technology of AI drawing is the same as AlphaGo, which is good at Go. They are all works that allow the system to deeply learn human beings. After absorbing a lot of data and knowledge, they finally achieve the goal of understanding human behavior. even emotional imitation. In the process of learning, how does AI ensure that the content it learns is reasonable, legal, and non-infringing, which has become the key to how AI creation can ensure the legality of its copyright?


Due to the abuse and unreasonable use of AI painting, the current capabilities of AI painting have attracted objections and protests from some groups of artists around the world. Polish conceptual artist Greg Rutkowski was concerned about the ability of AI to paint. Known for his magnificent oil painting style, he has recently become the most popular imitation object in AI painting. When he searched for his name on the Internet, all the AI ​​paintings popped up, but his own works were submerged. In a recent interview, Rutkowski lamented, “I feel like my career is under threat.”


After all, the computing power of AI is too powerful. Compared with AI that can spit out a painting that resembles Van Gogh’s expressionist style in three seconds, the creative efficiency that most artists take hours or even days to complete is indeed Incomparable. If the AI ​​mechanically copies, splices, and combines other people’s art, photography or graphic works through picture filters, etc. during the painting process, it will obviously constitute an infringement. For works that are not mechanically reproduced, it is necessary to further carry out the judgment of “substantial similarity”, decompose the paintings drawn by AI and the works to be compared, observe the similarities and differences between the two, and determine whether they constitute a Substantial infringement. In fact, the developers of drawing AI tools are generally the biggest beneficiaries of AI drawing and displaying their paintings to the outside world. For developers, while exploring AI creation, developers of AI painting tools should ensure the legality of the copyright source of painting materials, conduct necessary substantive similarity identification of imitative paintings before public publication, and update development in a timely manner It is an intelligent system upgrade to make up for algorithmic loopholes.


Every technological advancement will be accompanied by new technological ethics and social issues. Whether it is playing Go or painting, AI should still take into account its due social responsibilities and technological ethics in the process of pursuing its industrialization.

Reference:
Ren, H., Li, J., & Gao, N. (2019). Two-stage sketch colorization with color parsing. IEEE Access, 8, 44599-44610.
Park, T., Liu, M. Y., Wang, T. C., & Zhu, J. Y. (2019). GauGAN: semantic image synthesis with spatially adaptive normalization. In ACM SIGGRAPH 2019 Real-Time Live! (pp. 1-1).

Please rate this

What Can AI Do With Medical Image?

7

October

2022

No ratings yet.

The combination of medical image and artificial intelligence is a relatively new branch in the field of digital medical care (Savadjiev,2019), and it is also a hot spot in the digital medical industry. Medical images contain so much data that even experienced doctors can sometimes seem overwhelming. The interpretation of medical images requires the accumulation of professional experience over a long period of time, and the training period for radiologists is relatively long. However, artificial intelligence can detect images faster than professional doctors in efficiency and accuracy, reducing the human error rate.

Several well-known start-ups have emerged in many countries. Enlitic (Journal of Engineering, 2022) in the table is a relatively well-known artificial intelligence medical imaging company. Although it was founded in 2014, it was named one of the 50 smartest companies in the world in 2015 by MIT Technology Review in the following year and received a total of 15 million US dollars in financing. The Butterfly is developing a small ultrasound device that runs largely on software, including using techniques developed by artificial intelligence experts to comb through a series of images to extract features that can automatically diagnose disease. The world’s first and third most active venture capital institutions focusing on artificial intelligence have also become investors in smart medical imaging start-ups.

Judging from the number of misdiagnoses in imaging, the number of misdiagnoses in the United States has reached 12 million per year, while the number of misdiagnoses in China has reached an astonishing 57 million per year due to its huge population base. (Atencio, 2021) These misdiagnoses mainly occur in primary medical institutions. At present, medical imaging in China is transitioning from traditional film to electronic film, while traditional American film has become a thing of the past. The widespread use of electronic film has led to a substantial increase in medical imaging data, with an annual growth rate of 63.1% in the United States and 30% in China. The annual growth rate of radiologists in the United States and China is only 2.2% and 4.1%, which is far lower than the growth of imaging data, forming a huge gap. This means that the workload of doctors is greatly increased, and the accuracy of a judgment is reduced. The use of artificial intelligence to judge images can effectively fill the gap.

Reference:

“Hub Security Partners with Enlitic to Secure Health Data, Enhance Patient Care through Artificial Intelligence” (2022) Journal of Engineering, 829, pp. 829–829.

Atencio, Y. P. et al. (2021) “2021 International Conference on Electrical, Computer, Communications and Mechatronics Engineering (iceccme),” in Image Processing Techniques for Medical Applications. IEEE, pp. 1–6. doi: 10.1109/ICECCME52200.2021.9591045.

Savadjiev, P. et al. (2019) “Demystification of Ai-Driven Medical Image Interpretation: Past, Present and Future,” European radiology, 29(3), pp. 1616–1624. doi: 10.1007/s00330-018-5674-x.

Please rate this