Emerging Ethical Considerations for the Use of Artificial Intelligence in Ophthalmology

Rapid developments in artificial intelligence (AI) promise improved diagnosis and care for patients, but raise ethical issues.1Ting D.S.W. Pasquale L.R. Peng L. et al.Artificial intelligence and deep learning in ophthalmology.Br J Ophthalmol. 2019; 103: 167-175Crossref PubMed Scopus (411) Google Scholar, 2Lee A. Machine diagnosis.Nature. 2019 Apr 10; (Online ahead of print.)https://doi.org/10.1038/d41586-019-01112-xCrossref Google Scholar, 3Lin D. Lin H. Translating artificial intelligence into clinical practice.Ann Transl Med. 2020; 8: 715Crossref PubMed Google Scholar, 4Emanuel E.J. Wachter R.M. Artificial intelligence in health care: will the value match the hype?.JAMA. 2019; 321: 2281-2282Crossref PubMed Scopus (105) Google Scholar, 5Abràmoff M.D. Tobey D. Char D.S. Lessons learned about autonomous AI: finding a safe, efficacious, and ethical path through the development process.Am J Ophthalmol. 2020; 214: 134-142Abstract Full Text Full Text PDF PubMed Scopus (35) Google Scholar Over 6 months, in consultation with the American Academy of Ophthalmology Committee on Artificial Intelligence, we analyzed potential ethical concerns, with a focus on applications of AI in ophthalmology that are deployed or will be deployed in the near future.6Jobin A. Ienca M. Vayena E. The global landscape of AI ethics guidelines.Nat Mach Intell. 2019; 1: 389-399Crossref Google Scholar We identified 3 pressing issues: (1) transparency, paradigmatically through the explanation or interpretation of AI models; (2) attribution of responsibility issues for particular harms arising from the use or misuse of AI; and (3) scalability of use cases and screening infrastructure.TransparencyThe ability to understand why a machine learning model has produced a particular result is an oft-cited ethical principle for AI.4Emanuel E.J. Wachter R.M. Artificial intelligence in health care: will the value match the hype?.JAMA. 2019; 321: 2281-2282Crossref PubMed Scopus (105) Google Scholar, 5Abràmoff M.D. Tobey D. Char D.S. Lessons learned about autonomous AI: finding a safe, efficacious, and ethical path through the development process.Am J Ophthalmol. 2020; 214: 134-142Abstract Full Text Full Text PDF PubMed Scopus (35) Google Scholar7London A.J. Artificial intelligence and black-box medical decisions: accuracy versus explainability.Hastings Cent Rep. 2019; 49: 15-21Crossref PubMed Scopus (181) Google Scholar, 8Rudin C. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead.Nat Mach Intell. 2019; 1: 206-215Crossref PubMed Scopus (1489) Google Scholar, 9Gunning D. Explainable artificial intelligence (XAI). Paper presented at: DARPA/I20 program update; November 2017; Washington DC.Google Scholar, 10Hoffman R.R. Klein G. Mueller S.T. Explaining explanation for “explainable Ai.”.Proceedings of the Human Factors and Ergonomics Society Annual Meeting. 2018; 62: 197-201Crossref Google Scholar We distinguish between AI models that are interpretable, or governed by models that are directly understandable by humans, and AI models that are too complex for any human to comprehend (sometimes called “black box” models), requiring post hoc explainability for how results are produced.4Emanuel E.J. Wachter R.M. Artificial intelligence in health care: will the value match the hype?.JAMA. 2019; 321: 2281-2282Crossref PubMed Scopus (105) Google Scholar Recent work has shown that lack of is with accuracy of AI L. M. et ethical for a AI and and 2018; PubMed Scopus Google Scholar, deep with and on Scopus Google Scholar of for in and of D.S.W. Pasquale L.R. Peng L. et al.Artificial intelligence and deep learning in ophthalmology.Br J Ophthalmol. 2019; 103: 167-175Crossref PubMed Scopus (411) Google be an AI model or a a to for AI a on an or Machine learning and deep learning to be on of to of an are with a particular a is presented that is a AI model is an the AI model or transparency, be to why a particular the explanation is that the is are or be is to the for AI to and Machine learning in ophthalmology but to has improved M.D. M. et of an autonomous for of in care Med. 2018; 1: PubMed Scopus Google Scholar in a of interpretation of in in clinical a result of D. et interpretation of a 2017; Full Text Full Text PDF PubMed Scopus Google Scholar be to the use of AI to to is on the medical results be for be in the of of a C. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead.Nat Mach Intell. 2019; 1: 206-215Crossref PubMed Scopus (1489) Google Scholar ophthalmology is in with for to AI a in in the in are The to and how between and by of issues in the use of medical human on and to an AI are through from a of the an a but an is the be the diagnosis a AI models be to in to but lack in AI be for AI The for a to artificial medical Med. 2020; PubMed Scopus Google the of AI a of The clinical of screening is but is by and human D.S.W. Pasquale L.R. Peng L. et al.Artificial intelligence and deep learning in ophthalmology.Br J Ophthalmol. 2019; 103: 167-175Crossref PubMed Scopus (411) Google Scholar Artificial intelligence a screening in in and a the of for Scholar diagnosis to in The value of versus 2017; PubMed Scopus Google Scholar the potential and the use of AI of C. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead.Nat Mach Intell. 2019; 1: 206-215Crossref PubMed Scopus (1489) Google Scholar has that explainability be a the focus be on models that are interpretable, to for the a between AI or for the of to distinguish between the responsibility for AI in a and the or harms we with ethical and for are health a responsibility responsibility be to or health and and and has responsibility for in AI to the responsibility through responsibility for AI in by and on of the M.D. Tobey D. Char D.S. Lessons learned about autonomous AI: finding a safe, efficacious, and ethical path through the development process.Am J Ophthalmol. 2020; 214: 134-142Abstract Full Text Full Text PDF PubMed Scopus (35) Google are for that AI and but be for of the of for that for by the of the and M.D. Tobey D. Char D.S. Lessons learned about autonomous AI: finding a safe, efficacious, and ethical path through the development process.Am J Ophthalmol. 2020; 214: 134-142Abstract Full Text Full Text PDF PubMed Scopus (35) Google Scholar for use to to the but the of models between and AI use be in the to M.D. M. et of an autonomous for of in care Med. 2018; 1: PubMed Scopus Google Scholar, on in 2019; PubMed Scopus Google Scholar use on is issues in AI that of with a diagnosis in to responsibility for use are by the for and will the for or responsibility the of and is an that of AI focus on E. E. et of a deep learning deployed in for the of of the on Human Factors in for Scopus Google Scholar with a that AI of a of and and in the of that The for a to artificial medical Med. 2020; PubMed Scopus Google Scholar The to and harms is and the and of an AI in the that medical and clinical of human versus AI AI in a of to to potential and in the Google that by improved for the diagnosis of in AI and diagnosis for Scholar the that autonomous human M.D. M. et of an autonomous for of in care Med. 2018; 1: PubMed Scopus Google Scholar diagnosis for et artificial intelligence to medical in Transl Med. 2020; 8: PubMed Google Scholar of medical AI on in 2019; PubMed Scopus Google Scholar be to or of an the of and a for AI and promise of AI is to a the the from through for The for 2017; PubMed Scopus Google Scholar We a in AI an that cases of in the a for in a of for a that to a of of of of A. Machine diagnosis.Nature. 2019 Apr 10; (Online ahead of print.)https://doi.org/10.1038/d41586-019-01112-xCrossref Google Scholar to AI or results in a of for AI has with et in J Med. 2020; PubMed Scopus Google Scholar and A.J. et in medical for A. 2020; PubMed Scopus (105) Google Scholar in a of health in or are to from AI of a of and be by the H. et an to health Full Text Full Text PDF PubMed Scopus Google Scholar The to with AI applications is an we that in clinical is an in clinical to Scholar and into AI development the of harms AI be the of through the of of the of by AI; AI model is of to the potential harms of the of the for diagnosis or The ethics of 2020; Google Scholar, Pasquale L.R. in a J Ophthalmol. 2020; Full Text Full Text PDF PubMed Scopus Google Scholar Artificial to and of care for with et deep learning to 2020; Full Text Full Text PDF PubMed Scopus (3) Google Scholar and interpretation of potential between by AI a directly the and the of or cases a the of to a medical AI on to care a clinical medical ethical Med. 2017; 3 PubMed Google Scholar AI a AI be to the that or diagnosis is the of AI to in on the to of to be is by be by and on be to medical care in and through in medical AI is is a of the and is a that or for in the of a learning AI to with applications of learning machine 2020; Full Text Full Text PDF PubMed Scopus Google Scholar but through machine from and on how to the AI are in are to the from is to be through and will to and to the is but of the of results by human and the to on a the and medical of is but AI models the a to be directly to M. et to medical machine 2020; PubMed Scopus Google Scholar the of and but in and use AI in the and and in to the presented by the between the and the potential of AI in that the and is for the a be or through a for through in health artificial intelligence a of to medical care and to health care to the use of AI ethical concerns, in cases the of human and issues are in to and are in a of the and in is and the of AI will a on are is Rapid developments in artificial intelligence (AI) promise improved diagnosis and care for patients, but raise ethical issues.1Ting D.S.W. Pasquale L.R. Peng L. et al.Artificial intelligence and deep learning in ophthalmology.Br J Ophthalmol. 2019; 103: 167-175Crossref PubMed Scopus (411) Google Scholar, 2Lee A. Machine diagnosis.Nature. 2019 Apr 10; (Online ahead of print.)https://doi.org/10.1038/d41586-019-01112-xCrossref Google Scholar, 3Lin D. Lin H. Translating artificial intelligence into clinical practice.Ann Transl Med. 2020; 8: 715Crossref PubMed Google Scholar, 4Emanuel E.J. Wachter R.M. Artificial intelligence in health care: will the value match the hype?.JAMA. 2019; 321: 2281-2282Crossref PubMed Scopus (105) Google Scholar, 5Abràmoff M.D. Tobey D. Char D.S. Lessons learned about autonomous AI: finding a safe, efficacious, and ethical path through the development process.Am J Ophthalmol. 2020; 214: 134-142Abstract Full Text Full Text PDF PubMed Scopus (35) Google Scholar Over 6 months, in consultation with the American Academy of Ophthalmology Committee on Artificial Intelligence, we analyzed potential ethical concerns, with a focus on applications of AI in ophthalmology that are deployed or will be deployed in the near future.6Jobin A. Ienca M. Vayena E. The global landscape of AI ethics guidelines.Nat Mach Intell. 2019; 1: 389-399Crossref Google Scholar We identified 3 pressing issues: (1) transparency, paradigmatically through the explanation or interpretation of AI models; (2) attribution of responsibility issues for particular harms arising from the use or misuse of AI; and (3) scalability of use cases and screening ability to understand why a machine learning model has produced a particular result is an oft-cited ethical principle for AI.4Emanuel E.J. Wachter R.M. Artificial intelligence in health care: will the value match the hype?.JAMA. 2019; 321: 2281-2282Crossref PubMed Scopus (105) Google Scholar, 5Abràmoff M.D. Tobey D. Char D.S. Lessons learned about autonomous AI: finding a safe, efficacious, and ethical path through the development process.Am J Ophthalmol. 2020; 214: 134-142Abstract Full Text Full Text PDF PubMed Scopus (35) Google Scholar7London A.J. Artificial intelligence and black-box medical decisions: accuracy versus explainability.Hastings Cent Rep. 2019; 49: 15-21Crossref PubMed Scopus (181) Google Scholar, 8Rudin C. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead.Nat Mach Intell. 2019; 1: 206-215Crossref PubMed Scopus (1489) Google Scholar, 9Gunning D. Explainable artificial intelligence (XAI). Paper presented at: DARPA/I20 program update; November 2017; Washington DC.Google Scholar, 10Hoffman R.R. Klein G. Mueller S.T. Explaining explanation for “explainable Ai.”.Proceedings of the Human Factors and Ergonomics Society Annual Meeting. 2018; 62: 197-201Crossref Google Scholar We distinguish between AI models that are interpretable, or governed by models that are directly understandable by humans, and AI models that are too complex for any human to comprehend (sometimes called “black box” models), requiring post hoc explainability for how results are produced.4Emanuel E.J. Wachter R.M. Artificial intelligence in health care: will the value match the hype?.JAMA. 2019; 321: 2281-2282Crossref PubMed Scopus (105) Google Scholar Recent work has shown that lack of is with accuracy of AI L. M. et ethical for a AI and and 2018; PubMed Scopus Google Scholar, deep with and on Scopus Google Scholar of for in and of D.S.W. Pasquale L.R. Peng L. et al.Artificial intelligence and deep learning in ophthalmology.Br J Ophthalmol. 2019; 103: 167-175Crossref PubMed Scopus (411) Google be an AI model or a a to for AI a on an or Machine learning and deep learning to be on of to of an are with a particular a is presented that is a AI model is an the AI model or transparency, be to why a particular the explanation is that the is are or be is to the for AI to and Machine learning in ophthalmology but to has improved M.D. M. et of an autonomous for of in care Med. 2018; 1: PubMed Scopus Google Scholar in a of interpretation of in in clinical a result of D. et interpretation of a 2017; Full Text Full Text PDF PubMed Scopus Google Scholar be to the use of AI to to is on the medical results be for be in the of of a C. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead.Nat Mach Intell. 2019; 1: 206-215Crossref PubMed Scopus (1489) Google Scholar ophthalmology is in with for to AI a in in the in are The to and how between and by of issues in the use of medical human on and to an AI are through from a of the an a but an is the be the diagnosis a AI models be to in to but lack in AI be for AI The for a to artificial medical Med. 2020; PubMed Scopus Google the of AI a of The clinical of screening is but is by and human D.S.W. Pasquale L.R. Peng L. et al.Artificial intelligence and deep learning in ophthalmology.Br J Ophthalmol. 2019; 103: 167-175Crossref PubMed Scopus (411) Google Scholar Artificial intelligence a screening in in and a the of for Scholar diagnosis to in The value of versus 2017; PubMed Scopus Google Scholar the potential and the use of AI of C. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead.Nat Mach Intell. 2019; 1: 206-215Crossref PubMed Scopus (1489) Google Scholar has that explainability be a the focus be on models that are interpretable, to for the a between AI or for the of to The ability to understand why a machine learning model has produced a particular result is an oft-cited ethical principle for AI.4Emanuel E.J. Wachter R.M. Artificial intelligence in health care: will the value match the hype?.JAMA. 2019; 321: 2281-2282Crossref PubMed Scopus (105) Google Scholar, 5Abràmoff M.D. Tobey D. Char D.S. Lessons learned about autonomous AI: finding a safe, efficacious, and ethical path through the development process.Am J Ophthalmol. 2020; 214: 134-142Abstract Full Text Full Text PDF PubMed Scopus (35) Google Scholar7London A.J. Artificial intelligence and black-box medical decisions: accuracy versus explainability.Hastings Cent Rep. 2019; 49: 15-21Crossref PubMed Scopus (181) Google Scholar, 8Rudin C. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead.Nat Mach Intell. 2019; 1: 206-215Crossref PubMed Scopus (1489) Google Scholar, 9Gunning D. Explainable artificial intelligence (XAI). Paper presented at: DARPA/I20 program update; November 2017; Washington DC.Google Scholar, 10Hoffman R.R. Klein G. Mueller S.T. Explaining explanation for “explainable Ai.”.Proceedings of the Human Factors and Ergonomics Society Annual Meeting. 2018; 62: 197-201Crossref Google Scholar We distinguish between AI models that are interpretable, or governed by models that are directly understandable by humans, and AI models that are too complex for any human to comprehend (sometimes called “black box” models), requiring post hoc explainability for how results are produced.4Emanuel E.J. Wachter R.M. Artificial intelligence in health care: will the value match the hype?.JAMA. 2019; 321: 2281-2282Crossref PubMed Scopus (105) Google Scholar Recent work has shown that lack of is with accuracy of AI L. M. et ethical for a AI and and 2018; PubMed Scopus Google Scholar, deep with and on Scopus Google Scholar of for in and of D.S.W. Pasquale L.R. Peng L. et al.Artificial intelligence and deep learning in ophthalmology.Br J Ophthalmol. 2019; 103: 167-175Crossref PubMed Scopus (411) Google Scholar be an AI model or a a to for AI a on an or Machine learning and deep learning to be on of to of an are with a particular a is presented that is a AI model is an the AI model or transparency, be to why a particular the explanation is that the is are or be is to the for AI to and Machine learning in ophthalmology but to has improved M.D. M. et of an autonomous for of in care Med. 2018; 1: PubMed Scopus Google Scholar in a of interpretation of in in clinical a result of D. et interpretation of a 2017; Full Text Full Text PDF PubMed Scopus Google Scholar be to the use of AI to The to is on the medical results be for be in the of of a C. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead.Nat Mach Intell. 2019; 1: 206-215Crossref PubMed Scopus (1489) Google Scholar ophthalmology is in with for to AI a in in the in are The to and how between and by of issues in the use of medical human on and to an AI are through from a of the an a but an is the be the diagnosis a AI models be to in to but lack in AI be for AI The for a to artificial medical Med. 2020; PubMed Scopus Google Scholar the of AI a of The clinical of screening is but is by and human D.S.W. Pasquale L.R. Peng L. et al.Artificial intelligence and deep learning in ophthalmology.Br J Ophthalmol. 2019; 103: 167-175Crossref PubMed Scopus (411) Google Scholar Artificial intelligence a screening in in and a the of for Scholar diagnosis to in The value of versus 2017; PubMed Scopus Google Scholar the potential and the use of Explainable AI of C. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead.Nat Mach Intell. 2019; 1: 206-215Crossref PubMed Scopus (1489) Google Scholar has that explainability be a the focus be on models that are interpretable, to for the a between AI or for the of to distinguish between the responsibility for AI in a and the or harms we with ethical and for are health a responsibility responsibility be to or health and and and has responsibility for in AI to the responsibility through responsibility for AI in by and on of the M.D. Tobey D. Char D.S. Lessons learned about autonomous AI: finding a safe, efficacious, and ethical path through the development process.Am J Ophthalmol. 2020; 214: 134-142Abstract Full Text Full Text PDF PubMed Scopus (35) Google are for that AI and but be for of the of for that for by the of the and M.D. Tobey D. Char D.S. Lessons learned about autonomous AI: finding a safe, efficacious, and ethical path through the development process.Am J Ophthalmol. 2020; 214: 134-142Abstract Full Text Full Text PDF PubMed Scopus (35) Google Scholar for use to to the but the of models between and AI use be in the to M.D. M. et of an autonomous for of in care Med. 2018; 1: PubMed Scopus Google Scholar, on in 2019; PubMed Scopus Google Scholar use on is issues in AI that of with a diagnosis in to responsibility for use are by the for and will the for or responsibility the of and is an that of AI focus on E. E. et of a deep learning deployed in for the of of the on Human Factors in for Scopus Google Scholar with a that AI of a of and and in the of that The for a to artificial medical Med. 2020; PubMed Scopus Google Scholar The to and harms is and the and of an AI in the that medical and clinical of human versus AI AI in a of to to potential and in the Google that by improved for the diagnosis of in AI and diagnosis for Scholar the that autonomous human M.D. M. et of an autonomous for of in care Med. 2018; 1: PubMed Scopus Google Scholar diagnosis for et artificial intelligence to medical in Transl Med. 2020; 8: PubMed Google Scholar of medical AI on in 2019; PubMed Scopus Google Scholar be to or of an the of and a for AI distinguish between the responsibility for AI in a and the or harms we with ethical and for are health a responsibility responsibility be to or health and and and has responsibility for in AI to the responsibility through responsibility for AI in by and on of the M.D. Tobey D. Char D.S. Lessons learned about autonomous AI: finding a safe, efficacious, and ethical path through the development process.Am J Ophthalmol. 2020; 214: 134-142Abstract Full Text Full Text PDF PubMed Scopus (35) Google Scholar are for that AI and but be for of the of for that for by the of the and M.D. Tobey D. Char D.S. Lessons learned about autonomous AI: finding a safe, efficacious, and ethical path through the development process.Am J Ophthalmol. 2020; 214: 134-142Abstract Full Text Full Text PDF PubMed Scopus (35) Google Scholar for use to to the but the of models between and AI use be in the to M.D. M. et of an autonomous for of in care Med. 2018; 1: PubMed Scopus Google Scholar, on in 2019; PubMed Scopus Google Scholar use on is issues in AI that of with a diagnosis in to responsibility for use are by the for and will the for or responsibility the of and is an that of AI focus on E. E. et of a deep learning deployed in for the of of the on Human Factors in for Scopus Google Scholar with a that AI of a of and and in the of that The for a to artificial medical Med. 2020; PubMed Scopus Google Scholar The to and harms is and the and of an AI in the that medical and clinical of human versus AI AI in a of to to potential and in the Google that by improved for the diagnosis of in AI and diagnosis for Scholar the that autonomous human M.D. M. et of an autonomous for of in care Med. 2018; 1: PubMed Scopus Google Scholar diagnosis for et artificial intelligence to medical in Transl Med. 2020; 8: PubMed Google Scholar of medical AI on in 2019; PubMed Scopus Google Scholar be to or of an the of and a for AI and promise of AI is to a the the from through for The for 2017; PubMed Scopus Google Scholar We a in AI an that cases of in the a for in a of for a that to a of of of of A. Machine diagnosis.Nature. 2019 Apr 10; (Online ahead of print.)https://doi.org/10.1038/d41586-019-01112-xCrossref Google Scholar to AI or results in a of for AI has with et in J Med. 2020; PubMed Scopus Google Scholar and A.J. et in medical for A. 2020; PubMed Scopus (105) Google Scholar in a of health in or are to from AI of a of and be by the H. et an to health Full Text Full Text PDF PubMed Scopus Google Scholar The to with AI applications is an we that in clinical is an in clinical to Scholar and into AI development the of harms AI be the of through the of of the of by AI; AI model is of to the potential harms of the of the for diagnosis or The ethics of 2020; Google Scholar, Pasquale L.R. in a J Ophthalmol. 2020; Full Text Full Text PDF PubMed Scopus Google Scholar Artificial to and of care for with et deep learning to 2020; Full Text Full Text PDF PubMed Scopus (3) Google Scholar and interpretation of potential between by AI a directly the and the of or cases a the of to a medical AI on to care a clinical medical ethical Med. 2017; 3 PubMed Google Scholar AI a AI be to the that or diagnosis is the of AI to in on the to of to be is by be by and on be to medical care in and through in medical AI is is a of the and is a that or for in the of a learning AI to with applications of learning machine 2020; Full Text Full Text PDF PubMed Scopus Google Scholar but through machine from and on how to the AI are in are to the from is to be through and will to and to the is but of the of results by human and the to on a the and medical of is but AI models the a to be directly to M. et to medical machine 2020; PubMed Scopus Google Scholar the of and but in and use AI in the and and in to the presented by the between the and the potential of AI in that the and is for the a be or through a for through in health artificial intelligence a of to medical care and to health care to the use of AI ethical concerns, in cases the of human and issues are in to and are in a of the and in is and the of AI will a on are is promise of AI is to a the the from through for The for 2017; PubMed Scopus Google Scholar We a in AI an that cases of in the a for in a of for a that to a of of of of A. Machine diagnosis.Nature. 2019 Apr 10; (Online ahead of print.)https://doi.org/10.1038/d41586-019-01112-xCrossref Google Scholar to AI or results in a of for AI has with et in J Med. 2020; PubMed Scopus Google Scholar and A.J. et in medical for A. 2020; PubMed Scopus (105) Google Scholar in a of health in or are to from AI of a of and be by the H. et an to health Full Text Full Text PDF PubMed Scopus Google Scholar The to with AI applications is an we that in clinical is an in clinical to Scholar and into AI development the of harms AI be the of through the of of the of by AI; AI model is of to the potential harms of the of the for diagnosis or The ethics of 2020; Google Scholar, Pasquale L.R. in a J Ophthalmol. 2020; Full Text Full Text PDF PubMed Scopus Google Scholar Artificial to and of care for with et deep learning to 2020; Full Text Full Text PDF PubMed Scopus (3) Google Scholar and interpretation of potential between by AI a directly the and the of or cases a the of to a medical AI on to care a clinical medical ethical Med. 2017; 3 PubMed Google Scholar AI a AI be to the that or diagnosis is the of AI to in on the to of to be is by be by and on be to medical care in and through in medical AI is is a of the and The in medical AI is is a of the and The in medical AI is is a of the and is a that or for in the of a learning AI to with applications of learning machine 2020; Full Text Full Text PDF PubMed Scopus Google Scholar but through machine from and on how to the AI are in are to the from is to be through and will to and to the is but of the of results by human and the to on a the and medical of is but AI models the a to be directly to M. et to medical machine 2020; PubMed Scopus Google Scholar the of and but in and use AI in the and and in to the presented by the between the and the potential of AI in that the and is for the a be or through a for through in health artificial intelligence a of to medical care and to health care to the use of AI ethical concerns, in cases the of human and issues are in to and are in a of the and in is and the of AI will a on are is The the American Academy of Ophthalmology on Artificial Intelligence, and the American Academy of Ophthalmology Committee for and on