Update 🏠Home.py

🏠Home.py

@@ -39,7 +39,13 @@ st.markdown('''Defining the target variable is not only difficult; it can also h
 
 st.markdown('''These issues are not limited to hiring contexts. They arise in any case where there is no simple mathematical translation of a real-world problem. Take university admissions, for example. One might use an algorithm to predict which applicants will be “good” students.” What makes for a good student, though? A student who performs the best on exams at the end of their degree? A student who improves the most in their time at university? A student who doesn’t drop out, or who wins awards, or who gets a prestigious job after graduating, or contributes to the university in extracurricular activities? As with saying what makes for a good salesperson, the answer may be “some of everything,” and so again the question arises: how much of everything? Or consider another case: a news recommendation algorithm for an online platform. What makes for a “good” recommendation? Is it one that maximizes the user’s time on the platform, or that maximizes ad sales, or that is not “biased” along political lines (and then: which political lines?), or that best satisfies the user’s preferences, or that does not spread misinformation, or that prevents political polarization, or…? How these questions are answered—and how these different considerations are weighed against one another—has profound implications for fairness and other social and ethical concerns.''')
 
-st.markdown('''This point is vividly illustrated by how Google’s search engine handles results with misinformation. When designing a search engine, you want to give “good” search results. What makes for a good search result, though? A common answer is that good search results are relevant. But if a search engine gives search results that are simply relevant—without any other consideration taken into account—the search engine may spread misinformation [Phillips-Brown, manuscript](https://philarchive.org/archive/PHIANJ). Consider that as of December 12, 2016, Google’s search engine delivered the following results for the query “did the Holocaust happen?”''')
+st.markdown('''This point is vividly illustrated by how Google’s search engine handles results with misinformation. When designing a search engine, you want to give “good” search results. What makes for a good search result, though? A common answer is that good search results are relevant. But if a search engine gives search results that are simply relevant—without any other consideration taken into account—the search engine may spread misinformation [(Phillips-Brown, manuscript)](https://philarchive.org/archive/PHIANJ). Consider that as of December 12, 2016, Google’s search engine delivered the following results for the query “did the Holocaust happen?”''')
+
+st.image('./images/google_screenshot.png')
+
+st.markdown('''The top result here is from a neo-Nazi group—Stormfront—claiming that the Holocaust didn’t happen. Why was this a top result? Because the claim that the Holocaust didn’t happen, while false, is relevant to the question “did the Holocaust happen?” as Google noted [(Roberts 2016a)](https://fortune.com/2016/12/12/google-holocaust/). (As an analogy, imagine that you ask me “will it rain tonight?” I reply “yes,” but in fact it does not rain. I have said something that, while false, is relevant to your question.) Google has since changed its definition of what a “good” search result is [(Roberts 2016b)](https://fortune.com/2016/12/20/google-algorithm-update/). Now, a “good” search result for Google is not a result that is simply relevant, but that is also true. Today, results for the query “did the Holocaust happen?” do not include Holocaust denial. For example, the first result for this query is instead:''')
+
+st.image('./images/google_screenshot_2.png')
 
 st.markdown('''Target variable definition, then, is not a merely technical matter. The question of what makes for a “good” employee, student, or news recommendation (and so on) is fundamentally value-laden. It calls for close attention and transparency [(Fazelpour & Danks, 2021)](https://compass.onlinelibrary.wiley.com/doi/full/10.1111/phc3.12760). All too often, though, target variables are defined in technical settings without attention to fairness. Further, stakeholders who aren't a part of the technical process—like managers in non-technical roles, or those working in upper management or human resources—either do not understand, or are simply not aware of, the fraught nature of target variable definition.''')