A reinforcement learning model developed to adapt Artificial Intelligence (AI) predictions to human preferences showed better performance for skin cancer diagnoses and optimal management decisions compared to a supervised learning model HaraldKittler
, lowering the threshold results in a high number of patients with >3 excised benign lesions and with an increase of excised melanomas . The RL model would remove 61.8% and monitor 20% of melanomas, outperforming all other models in terms of acceptable management decisions for these melanomas . At the same time, 23 patients would have >3 benign lesions removed. A distinctive feature of the RL model would be the high number of benign lesions .
Here, we demonstrate that the integration of human preferences, represented as reward tables created by experts, enhances the performance of a pretrained AI decision-support system. Improvement is evident in both the system’s standalone performance and its ability to collaborate effectively with dermatologists. Dermatologists’ improvement may be due to the RL model reducing AI overconfidence by considering consequences of management decisions.
Based on our results, we suggest that RL, among other techniques, could be a suitable tool for this purpose, although it is not necessarily the best solution. A limitation of the RL method is that the model must be retrained, whereas other simpler approaches, such as thresholding, can be applied without retraining.
In conclusion, our study shows that incorporating human preferences can improve AI-based diagnostic decision support and that such preferences could be considered when developing AI tools for clinical practice. RL could be a potential alternative to threshold-based methods for creating tailored approaches in complex clinical scenarios. However, additional research, including evaluating patient and provider satisfaction, is necessary to fully uncover the potential of RL in this context.
Malaysia Latest News, Malaysia Headlines
Similar News:You can also read news stories similar to this one that we have collected from other news sources.
Hands On With Google Search’s Answer to ChatGPTGoogle’s new AI-powered search engine can feel more like artificial interference than artificial intelligence.
Read more »
Correcting the drug development paradigm for glioblastoma requires serial tissue sampling - Nature MedicineCorrecting the drug development paradigm for glioblastoma requires serial tissue sampling. Comment from MKhasraw and colleagues, from the 27th Society of Neuro-Oncology annual meeting.
Read more »
If artificial intelligence takes your job, AI should pay you, argues actor''Generative AI' cannot generate anything at all without first being trained on massive troves of data it then recombines. Who produces that training data? People do. And those people deserve residuals,' argues actor Joseph Gordon-Levitt in a new Washington Post column. Gordon-Levitt joins Morning Joe to discuss.
Read more »
Opinion | If artificial intelligence uses your work, it should pay youOpinion by Joseph Gordon-Levitt: People whose jobs are threatened by AI will be the same people who produced the data used to train it. A new kind of residuals for these human data producers could potentially provide some much-needed economic relief.
Read more »
Artificial intelligence can detect 'sextortion' before it happens and help FBI: expertAI powers company's fight against sextortion scams, detects real or 'deepfake' sexually explicit images
Read more »
What Alphabet and Microsoft are saying about A.I. on their earnings callsSecond-quarter earnings season brought artificial intelligence frontrunners Alphabet and Microsoft head to head one again.
Read more »